1S-2S Spectroscopy of Trapped Hydrogen: The

Cold Collision Frequency Shift and Studies of BEC

by

Thomas Charles Killian

Submitted to the Department of Physics
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy

at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
February 1999
© Massachusetts Institute of Technology 1999. All rights reserved.

A uthor ....................... ... . .... ..........

Department of Physics
January 11, 1999

Certified by.................................. ........

Thomas J. eytak
Professor o hysics
Thesis Supervisor
Certified by ........................
Dani'i Kleppner
Lester Wolfe Professor of Physics
Thesis Supervisor
aI
Accepted by .......... .....................P .. .- ......... .-- - -
. . . . . .

MASSAEHUSETTS INSTITUTE T! mas J. reytak

I n, Department of Physics Graduate ommittee
- ..

gMOFTECHNO

113R-ARIES
 1S-2S Spectroscopy of Trapped Hydrogen: The Cold
Collision Frequency Shift and Studies of BEC
by
Thomas Charles Killian

Submitted to the Department of Physics

on January 11, 1999, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy

Abstract
The cold collision frequency shift of the 1S-2S two-photon transition is studied in
trapped spin-polarized atomic hydrogen at submillikelvin temperatures. This effect
is the low temperature manifestation of the pressure shift and broadening familiar
from spectroscopy at normal temperatures and pressures. We find the shift is given
by Avs-2 s= -3.8 ± 0.8 x 10-10 n Hz cm 3 , where n is the sample density.
Theory is developed to express the shift in terms of the mean field interaction
energy due to collisions and thus relate it to the s-wave triplet scattering lengths,
ais-is and aIs-2s. From this we derive aIs-2s = -1.4 ± 0.3 nm, which is in fair
agreement with a recent calculation.
1S-2S spectroscopy is a valuable probe of the density, temperature, and atom-
atom interactions in the trapped sample, especially in the regime of Bose-Einstein
condensation (BEC). We describe properties of the condensate and how they are
determined from the 1S-2S spectrum.

Thesis Supervisor: Thomas J. Greytak

Title: Professor of Physics

Thesis Supervisor: Daniel Kleppner

Title: Lester Wolfe Professor of Physics
Acknowledgments
Graduate school has been the best time of my life and I owe that to Amy Salzhauer.
She always shares my excitement when things go well and is ready with a reassuring
smile when progress seems slow or nonexistent. She has a talent for getting me out
of the lab to enjoy the rest of life, and my experience at MIT has been richer because
I have had someone with which to share it.
For as long as I can remember my parents have encouraged me to ask questions,
take risks, and explore the world around me, and those lessons have served me well.
My family may not have understood why I spent the last five years staying up all
night working in the lab, but they were always supportive. Thanks!
I couldn't have asked for better advisors than Tom Greytak and Dan Kleppner.
They give their students the freedom to grow and take on challenges, and manage to
encourage hard work without forgetting that above all, physics is fun.
In an experiment like this one, nothing gets done unless people can work together
as a team, and this thesis wouldn't exist without the help of others in the lab.
I have learned a tremendous amount from working with Dale Fried, and I feel safe
in saying that with him more than anyone else, I share feelings of great satisfaction
and joy (and relief!) from finally seeing the experiment deliver on its promises.
Lorenz Willmann brought experience, a good sense of humor, and a passion for
rigorous data analysis to the experiment - all of which were desperately needed. It
is always a pleasure to work with him, even when learning about the subtleties of
chi-squared.
I am envious of the younger graduate students, Dave Landhuis and Stephen Moss,
because I am sure they will accomplish and learn amazing things in the next few years.
It is hard to leave something that has absorbed so much of my sweat and tears, but
it is made much easier by the talent and remarkable good nature of the people who
will take care of it.
I would have been lost in the lab if Claudio Cesar hadn't been here when I arrived.
He taught me all that I know about laser physics, and we had a great time in the
process.
It has been a privilege to work and share an office with Adam Polcyn. I have

never met anyone with a better disposition or more patience, and with all the phone

messages he took for me, I am sure I tested the latter!

Some of the most rewarding experiences I had at MIT have been working with

undergraduates. In particular I value my friendships with Jonathan Goldman and

Sourav Mandal and the chance I had to watch them learn about physics.
I also thank Ian Applebaum for help with numerical computations and Professors
Wolfgang Ketterle and Leonya Levitov for many useful discussions.
Contents

1 Introduction 15

2 Overview of Cryogenic Trapping and Cooling of Atomic Hydrogen 18

2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Trapping Low Field Seeking Hydrogen Atoms . . . . . . . . . . . . . 20
2.2.1 M agnetic Trap . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.2 Forming Atomic Hydrogen in a Radio Frequency Discharge . . 22
2.2.3 Spin Polarization . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.4 Loss Processes and Gas-Surface Equilibrium . . . . . . . . . . 24
2.3 Evaporative Cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Forced Evaporative Cooling . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.1 Magnetic Field Saddlepoint Evaporation . . . . . . . . . . . . 28
2.4.2 Radio-Frequency Evaporation . . . . . . . . . . . . . . . . . . 30
2.5 Probing the Trapped Gas . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5.1 Bolometric Temperature Measurement . . . . . . . . . . . . . 31
2.5.2 Bolometric Density Measurement . . . . . . . . . . . . . . . . 32

3 Overview of 1S-2S Spectroscopy in a Trap, 35

3.1 Background ..... ........................ 35
3.2 Two-Photon Excitation . . . . . . . . . . . . . . . . . . 36
3.3 Detection Scheme . . . . . . . . . . . . . . . . . . . . . 37
3.4 Microchannel Plate Photon Counter . . . . . . . . . . . 40
3.5 Laser System . . . . . . . . . . . . . . . . . . . . . . . 41

5
 3.6 Photoexcitation Spectrum . . . . . . . . . . . . . . . . . . . . . . . .4 43

4 Formal Description of 1S-2S Two-Photon Spectroscopy 46

4.1 15-2S Two-Photon Transition Theory . . . . . . . . . . . . . . . . . 46
4.1.1 Physical System and Interaction . . . . . . . . . . . . . . . . . 46
4.1.2 Excitation Hamiltonian . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Doppler-Sensitive Excitation . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Doppler-Free Excitation: Simple Lineshapes . . . . . . . . . . . . . . 53
4.3.1 Atoms Nearly at Rest . . . . . . . . . . . . . . . . . . . . . . 53
4.3.2 Atoms in Motion: Time-of-Flight Lineshape . . . . . . . . . . 54
4.4 Doppler-Free Excitation: Numerical Simulation of Complicated Line-
shap es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4.1 Effective Two-Level Hamiltonian and Evolution of the Single
Atom Density Matrix . . . . . . . . . . . . . . . . . . . . . . . 61
4.4.2 Cold Collision Frequency Shift . . . . . . . . . . . . . . . . . . 62
4.4.3 Additional Sources of Spectral Broadening . . . . . . . . . . . 63
4.4.4 Numerical Calculation of the Time of Flight Lineshape . . . . 66

4.4.5 Coherence Effects . . . . . . . . . . . . . . . . . . . . . . . . . 68

5 Cold Collision Frequency Shift: Observations 72

5.1 D ata . . . . . . . . . . . . . .. . . . .. . . . . . . . . . . . . . . . . . 72
5.1.1 Experimental Procedure . . . . . . . . . . . . . . . . . . . . . 72
5.1.2 D ata Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.1.3 Inhomogeneous Broadening and Shift . . . . . . . . . . . . . . 76
5.1.4 Systematic Uncertainties . . . . . . . . . . . . . . . . . . . . . 77
5.1.5 Measured Value of x . . . . . . . . . . . . . . . . . . . . . . . 78
5.2 The 15-25 S-Wave Triplet Scattering Length . . . . . . . . . . . . . 79
5.2.1 Experimental Value of the 1S-2S Scattering Length . . . . . . 79
5.2.2 Comparison with Theory . . . . . . . . . . . . . . . . . . . . . 80
5.3 Using the Cold Collision Frequency Shift as a Probe of the Trapped Gas 80
5.3.1 Noncondensed Gas . . . . . . . . . . . . . . . . . . . . . . . . 80

6
 5.3.2 Bose-Einstein Condensation . . . . . . . . . . . . . . . . . . . 83

6 Cold Collision Frequency Shift: Mean Field Theory 84

6.1 Mean Field Description for a Spatially Homogeneous System . . . . . 86
6.1.1 State Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.1.2 Hamiltonian and Mean Field Energies . . . . . . . . . . . . . 89
6.1.3 D iscussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Mean Field Description for a Bose Condensed Gas in a Magnetic Trap 93
6.2.1 System before Excitation . . . . . . . . . . . . . . . . . . . . . 94
6.2.2 System after Excitation . . . . . . . . . . . . . . . . . . . . . 95
6.2.3 D iscussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3 Cold Collision Frequency Shift for an Arbitrary System . . . . . . . . 102
6.3.1 Sum Rule for the Mean Frequency Shift in the Spectrum . . . 102
6.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

7 Spectroscopic Studies of a Quantum Degenerate Hydrogen Gas 110

7.1 Spectroscopic Signature of Bose-Einstein Condensation . . . . . . . . 111
7.2 Doppler-Free Spectrum of the Condensate . . . . . . . . . . . . . . . 112
7.3 Condensate Fraction . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.3.1 Theoretical Limit of the Condensate Fraction in Hydrogen . . 113
7.3.2 Spectroscopic Determination of the Condensate Fraction . . . 114
7.3.3 Determination of the Condensate Fraction from the Peak Shift
in the Doppler-Free Spectrum . . . . . . . . . . . . . . . . . . 119
7.3.4 Implications of the Measurement of the Condensate Fraction . 119
7.4 Phase Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.5 Spectrum of the Normal Fraction in the Degenerate Regime . . . . . 123
7.6 1S-2S Spectroscopy as a Probe of BEC . . . . . . . . . . . . . . . . . 125

8 Future Prospects 126

A High Resolution Spectroscopy 129

A.1 Best Achieved Resolution . . . . . . . . . . . . . . . . . . . . . . . . . 129

7
 A.1.1 Time-of-Flight Broadened Lines . . . . . . . . . . . . . . . . . 129
A.1.2 Coherence Effects . . . . . . . . . . . . . . . . . . . . . . . . . 131
A.2 Laser Frequency Stability Limitations . . . . . . . . . . . . . . . . . . 131
A.2.1 Reference Cavity Shift with Light Power . . . . . . . . . . . . 131
A.2.2 Doppler-Shifts Along the Beam Path . . . . . . . . . . . . . . 134
A.3 Prospects for Improving the Frequency Stability . . . . . . . . . . . 137

B 1S-2S Spectroscopy Appendix 138

B.1 Effective Two-Level Hamiltonian . . . . . . . . . . . . . . . . . . . . 138
B.2 Numerical Calculation of the Spectrum . . . . . . . . . . . . . . . . . 141

C Boltzmann Transport Equation Derivation of the Cold Collision Fre-

quency Shift 145
C.1 Evolution of the Single Atom Density Matrix . . . . . . . . . . . . . . 145
C.2 Quantum Boltzmann Transport Equation . . . . . . . . . . . . . . . . 147
C.3 Application to the 1S-2S Transition in Trapped Hydrogen . . . . . . 147
C .4 D iscussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

D Details of the Mean Field Theory Calculation of the Cold Collision

Frequency Shift 150
D.1 Correlation Functions for a Homogeneous System . . . . . . . . . . . 150
D.2 Interaction Energy for a Homogeneous System before Excitation . . . 152
D.3 Interaction Energy for a Homogeneous System after Excitation . . . . 152

D.4 Derivation of the Energy Functional for the Excited State of a Con-
densed Gas in a Trap . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
D.5 Details of Elements of the Proof of the Sum Rule for the Mean Fre-
quency of the Spectrum for an Arbitrary System . . . . . . . . . . . . 157

E 130 Te Reference Spectroscopy 161

2

E .1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
E.2 Saturated Absorption Spectroscopy . . . . . . . . . . . . . . . . . . . 163
E.2.1 Laser System . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

8
 E.2.2 1S-2S F = 1 Reference Transition, i 2 . . . . . . . . . . . . . . 165

E.3 Systematics of i 2 Frequency Stability . . . . . . . . . . . . . . . . . . 167

E.3.1 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
E.3.2 Column Density . . . . . . . . . . . . . . . . . . . . . . . . . . 168
E.3.3 Reliability and Cell to Cell Variation . . . . . . . . . . . . . . 168
E.4 Details of MIT Experimental Procedure . . . . . . . . . . . . . . . . 170

9
List of Figures

2-1 Hyperfine structure of the iS ground state of hydrogen in a magnetic

field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2-2 Overview of the trapping apparatus . . . . . . . . . . . . . . . . . . . 21
2-3 Diagram of the discharge and cell top . . . . . . . . . . . . . . . . . . 23
2-4 Schematic of the magnetic trap and hydrogen cloud shortly after load-
ing the trap ........ ................................ 26
2-5 Time between collisions, T = 1/na'iiv/2 for various hydrogen densities
and tem peratures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2-6 Zeeman diagram of IS F = 1 states in low magnetic fields showing the
transitions driven for RF evaporation . . . . . . . . . . . . . . . . . . 29
2-7 Bolometric determination of the sample temperature . . . . . . . . . 31
2-8 Decay of a trapped hydrogen sample . . . . . . . . . . . . . . . . . . 32

3-1 Level scheme for 1S-2S spectroscopy of magnetically trapped atomic

hydrogen......... .................................. 37
3-2 Excitation and detection . . . . . . . . . . . . . . . . . . . . . . . . . 38
3-3 Typical timing sequence for 15-2S excitation and detection . . . . . . 39
3-4 Laser system for 1S-2S spectroscopy of trapped atomic hydrogen . . 41
3-5 Composite 15-25 two-photon spectrum of trapped hydrogen . . . . . 45

4-1 Doppler-sensitive excitation spectrum of a sample held at a trap depth

of 280 p K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4-2 Cross sections of the laser beam and the trajectory of an atom in the
x - y plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

10
4-3 Typical Doppler-free spectra showing the dependence of linewidth on
sam ple tem perature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4-4 Doppler-free spectra from cold, low density samples: 1/e half linewidths
and inferred temperatures. . . . . . . . . . . . . . . . . . . . . . . . . 59
4-5 Probability for excitation to the 2S state for atoms which start outside
the laser beam and make one pass through the beam with zero impact
param eter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4-6 Contribution to the excitation rate from all the atoms with a given
atomic velocity in the laser beam, for various detunings from resonance 67
4-7 Numerical calculation of the probability for excitation to the 2S state
after a 1 m s laser pulse . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4-8 Spectrum of a 275 pK sample showing coherence sidebands . . . . . . 70

5-1 Doppler-free spectra of a 120 pK sample with an initial peak sample

density of no = 5.0 x 1013 CM-3 . . . . . . . . . . . . . . . . . . . . . 73

5-2 Analysis of spectra from a 120 [tK sample with an initial density of
6.6 x 1013 cm -3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5-3 The frequency shift parameter X, determined as described in the text
for various trap configurations . . . . . . . . . . . . . . . . . . . . . . 78
5-4 Spectra of a 55 [pK sample (trap depth = 350 pK) with initial peak
sample density of between (1 ~ 2) x 10 CM~3 . . . . . . . . . . . . . 81

6-1 Doppler-free spectra of noncondensed trapped hydrogen, showing the

cold collision frequency shift . . . . . . . . . . . . . . . . . . . . . . . 85
6-2 Probability distributions of the number of atoms in the 2S state for a
total of 10 7 atoms and various fractions of atoms excited . . . . . . . 89
6-3 1S-2S energy level diagram for a noncondensed, homogeneous sample,
showing the density-dependent level shift which gives rise to the cold
collision frequency shift . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6-4 Radial potentials and single particle wave functions for IS atoms in
the condensate and 2S atoms trapped in the condensate interaction well 96

11
6-5 Theoretical Doppler-free spectrum of a condensate at T = 0 in a three-
dimensional harmonic trap . . . . . . . . . . . . . . . . . . . . . . . . 100
6-6 Doppler-free spectrum of a condensate: comparison of theory and ex-
perim ent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6-7 Graphical depiction of the sum rule . . . . . . . . . . . . . . . . . . . 103
6-8 Excitation of the system to excited state ivi) . . . . . . . . . . . . . . 105

7-1 Composite 1S-2S two-photon spectrum of trapped hydrogen . . . . . 111

7-2 Typical laser and sample dimensions at the BEC transition . . . . . . 115
7-3 Doppler-free spectra of normal fraction and condensate . . . . . . . . 117
7-4 Spectroscopic determination of the condensate fraction . . . . . . . . 118
7-5 Time evolution of the Doppler-free spectrum of a single condensate . 121
7-6 BEC phase diagram of hydrogen and a typical evaporative cooling path 123
7-7 Doppler-free spectrum of the normal fraction immediately after the
end of the forced evaporation . . . . . . . . . . . . . . . . . . . . . . 124

A-i Spectroscopy of cold (<40 piK) low density (<1013 cm- 3 ) hydrogen . 130
A-2 Spectra showing motional sidebands . . . . . . . . . . . . . . . . . . . 132
A-3 Reference cavity transmitted power as measured by the FND-100Q
photodiode/amplifier circuit . . . . . . . . . . . . . . . . . . . . . . . 133
A-4 Response of the reference cavity mode frequency to a sudden change
in light level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
A-5 Schematic of the Doppler servo system . . . . . . . . . . . . . . . . . 135
A-6 Beatnote of Doppler servo system . . . . . . . . . . . . . . . . . . . . 136

B-i Energy diagram for radial motion of an atom in the magnetic trap
including the centrifugal potential, L 2 /2mr 2 . . . . . . . . . . . . . . 142

E-i Doppler-sensitive and saturated absorption spectrum of 13 0 Te

2 . . . . 162
E-2 Components of the laser system which are important for Tellurium
spectroscopy....... ................................ 164

12
E-3 Saturated absorption spectrum near 1/4 of the hydrogen 1S-2S F = 1
transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
13 0
E-4 High resolution saturated absorption spectrum of the 12 line in Te 2 166
13 0
E-5 Temperature dependence of the frequency of the i2 line in Te2 at

saturated vapor pressure . . . . . . . . . . . . . . . . . . . . . . . . . 167

E-6 Absorption on the center of the Doppler-broadened line 1284 . . . . . 169
E-7 Frequency of i 2 plotted against linewidth . . . . . . . . . . . . . . . . 169
E-8 Linewidth of i 2 versus oven temperature . . . . . . . . . . . . . . . . 170
E-9 Frequency of i 2 observed in two other cells, plotted against linewidth 171

13
List of Tables

5.1 Data for determining the cold collision frequency shift . . . . . . . . . 77

14
Chapter 1

Introduction

Spectroscopy of atomic hydrogen has contributed to some of the major advances in

physics in the twentieth century. Bohr's model to explain the hydrogenic electronic
spectrum [1] was the bridge from classical to quantum mechanics. The fine structure
splitting of the Balmer-alpha 2P-3S,3D transition led Sommerfeld to incorporate
relativity in the description of the atom [2], a program which was continued by Dirac
with his relativistic quantum theory [3]. Rabi's measurement of the ground state
hyperfine structure [4, 5] suggested the existence of the anomalous electron magnetic
moment. That work, along with Lamb's discovery of the 2S Lamb shift [6, 7], spurred
the development of quantum electro-dynamics.
Today, hydrogen spectroscopy continues to occupy a prominent position in physics
research, and there is particular interest in the two-photon 1S-2S transition because
of its narrow natural linewidth, 1.31 Hz, at a resonance frequency of 2.46 x 1015 Hz.
The 1S-2S transition, excited in an atomic beam, is part of the most precise measure-
ment of the IS Lamb shift, which now stretches our understanding of the structure
of the proton and quantum chromodynamics[8]. The transition is also important for
metrology and fundamental physical measurements. It has been suggested as a fre-
quency reference[9], and is used in the most precise determination of the Rydberg
constant[9].
This thesis describes new applications for 15-2S hydrogen spectroscopy. We study
H-H interactions in submillikelvin, magnetically trapped hydrogen[10] through the

15
cold collision frequency shift of the transition frequency [11, 12]. In addition, we
show how 1S-2S spectroscopy and the cold collision frequency shift can be used to
investigate Bose-Einstein condensation (BEC) in hydrogen[13].
The cold collision frequency shift is the low temperature manifestation of the
pressure shift and broadening familiar from spectroscopy at normal temperatures and
pressures [14]. In cold collisions, the temperature is so low that only a single partial
wave contributes to atom-atom scattering. Collisional frequency shifts in this regime
have been studied in the microwave region because they limit the accuracy of the
cryogenic hydrogen maser [15] and atomic fountain clocks based on cesium[16, 17, 18].
The 1S-2S observations described here extend this research from the microwave to
the optical region.
In hydrogen, the shift can be related to the s-wave elastic triplet scattering lengths
for 1S-2S and IS-IS collisions, as-2s and ais-is. The IS-IS scattering length is
known accurately from theory. To our knowledge these results constitute the first
measurement of a scattering length involving an excited state, and it tests the under-
standing of H-H molecular potentials.
The theory for the cold collision frequency shift in masers and fountain clocks has
been thoroughly developed [15], but this theory strictly applies only to a homogeneous
noncondensed system. The extension to an inhomogeneous gas, and to BEC, is not
trivial. The frequency shift is sensitive to atom-atom spatial correlations, and raises
interesting new questions about the state of the system after excitation.
The study of dilute degenerate quantum gases has captured the attention of the
physics community and the popular press since the condensation of rubidium [19],
sodium [20], and lithium [21, 22] in 1995. The observation of BEC in hydrogen capped
a 22-year research effort, and 1S-2S spectroscopy was an essential tool for detecting
the phase transition. High resolution spectroscopy is a new method for studying BEC
and it opens another window into this exciting phenomenon.
The techniques for magnetically trapping and cooling hydrogen are discussed in
Chap. 2. The experimental details of 1S-2S spectroscopy are described in Chap. 3.
Chapter 4 provides a theoretical description of two-photon excitation in a trap and

16
presents spectra observed under various experimental conditions.
Chapter 5 describes the data and analysis for the cold collision frequency shift
measurements, and Chap. 6 discusses the theory for the shift, which is necessary for
relating the shift to the s-wave scattering lengths. Chapter 7 discusses some of the
BEC measurements that can be extracted from the 1S-2S spectrum, and Chap. 8
describes future prospects for the experiment.
Appendix A describes the current limitations and future potential of high reso-
lution 1S-2S spectroscopy in the magnetic trap. Appendices B-D provide detailed
derivations of many of the calculated results presented in the bulk of the thesis, and
App. E describes the " 0Te2 reference spectrometer used to locate the frequency of
the 1S-2S transition.
This work should be viewed as a companion to D. Fried's Ph.D. thesis [23], "Bose-
Einstein Condensation of Atomic Hydrogen," which discusses the condensate data and
properties in much greater detail. This experiment is a group effort, and the emphasis
here is on describing how the 1S - 2S transition is used to study the trapped gas,
and on pointing out new spectroscopic effects.

17
Chapter 2

Overview of Cryogenic Trapping

and Cooling of Atomic Hydrogen

The study of magnetically trapped atomic hydrogen has been motivated by the pur-
suit of BEC, a phase transition to a state in which a macroscopic number of atoms
occupies the lowest energy level of the system. The transition occurs at low temper-
atures and high densities when noA ~ 2.612, where no is the peak sample density,
AT =h/ /2w7mkBT is the thermal de Broglie wavelength, and noA4 is the peak phase
space density in the sample.
The techniques for trapping and cooling hydrogen have been discussed extensively
in the literature (references given below), and also in the Ph.D. thesis of J. Doyle [24].
This chapter gives an overview of this aspect of the experiment and a bit of historical
background.

2.1 Background
The study of gaseous atomic hydrogen in the quantum regime can be traced back
to Hecht[25], who in 1959 pointed out that in a strong magnetic field the system
would remain a gas to T = 0 and, at low enough temperatures, might display the
effects of quantum degeneracy and superfluidity. There was little initial interest, but
experimental work started in earnest in the 1970's after further details were worked

18
 I I I I I I

400 d
cHt
-C
2001
d=ltt>
E
F=1 c=cos6It4>+sin8I4t >
m
0

wD F=O a=cosOlt4>-sinelJt >

-
-200
b
a
-400

0 0.12 0.24 0.36 0.48 0.60

Magnetic Field [Tesla]

Figure 2-1: Hyperfine structure of the IS ground state of hydrogen in a magnetic

field. The symbol Itt) denotes the state with the electron spin up, me = +1/2, and
the proton spin up, mp = +1/2, etc. The mixing angle is defined by tan(20) =
A/h('ye + 'yp)B = .0506/B, where A/h is the zero field hyperfine splitting, and Ke
are 7, are the electron and proton gyromagnetic ratios, respectively. In high field
(B > 0.05 T), the electron and proton spins couple to the magnetic field; in low field
the hyperfine coupling dominates. Electron spin up atoms (Ht) are pulled towards
the magnetic field minimum, while the electron spin down atoms (H) are pulled
towards the maximum of the field. According to convention, states are labelled a
through d in order of increasing energy.

out by Stwalley and Nosanow[26] and others [27]. To stabilize the system against
molecular recombination, a magnetic field is required to spatially separate atoms
with different electron spins. This greatly suppresses the exothermic reaction H + H
=> H2 + 4.6 eV, because electron spin-polarized atoms interact through the repulsive
triplet molecular potential[28, 29].
Atomic hydrogen interacts with a magnetic field through the well known Zeeman
effect [30]. The orientation of an atom's spin tends to follow the field adiabatically
due to the spin angular momentum, so the magnetic potential, U = -p - B, reduces
to a function of the magnitude of the field, as shown in Fig. 2-1. Atoms with their

19
electron spin down (H4, states a and b) are pulled towards a magnetic field maximum.
Atoms with their electron spin up (HT, states c and d) are pulled towards a magnetic
field minimum.
H4 atoms were magnetically stabilized by Silvera and Walraven [31] in 1979. The
sample was confined at 300 mK in a liquid 4 He coated cell in fields of up to 7 T. In
a similar apparatus, nuclear polarization was demonstrated in 1981 when the group
of Greytak and Kleppner produced a sample of b state atoms [32].
Bose condensing high field seeking atoms proved impossible in these experiments
because the trapped H4 atoms were always in thermal and diffusive contact with
containment walls. This limited the temperature to greater than about 100 mK and
three body recombination limited the sample densities to less than ~ 1018 cm- 3 [33],
below the BEC critical density. Containment walls were essential because Maxwell's
equations forbid the existence of a static magnetic field maximum in free space.
Hess [34] suggested wall free confinement of low field seekers in a magnetic field
minimum, and evaporative cooling, as a path towards BEC. This led to the develop-
ment of the current MIT hydrogen experiment.

2.2 Trapping Low Field Seeking Hydrogen Atoms

2.2.1 Magnetic Trap

For the experiments described in this thesis, a loffe-Pritchard [35] magnetic trap for
HT atoms is produced by superconducting magnets[24]. Large solenoids provide axial
confinement, and radial confinement is provided by a quadrupole magnetic field whose
magnitude increases linearly with r, the distance from the center axis, with gradient
B'. The field profile on axis is shown in Fig. 2-2. The potential seen by d state atoms
near the field minimum is U(r) ~ iB/B (1 + z 2 /z2) + (rB')2 , where PB is the Bohr
magneton (PB/h = 14 GHz/T, PB/kB= 0.67 K/T). The minimum field in the trap
is denoted B 0 . It is important for future discussions to note that the curvature along
the z axis is weak (Bo/zo < B').

20
 Dilution
Refrigerator
80
Cryogenic
Discharge
60
Trapped
Atoms
(H T)
I 40
Liquid
Helium
Coated
Cell 20

B (Tesla) N

4 3 2 1
Figure 2-2: Overview of the trapping apparatus. The cylindrically symmetric trap-
ping cell is thermally connected to a dilution refrigerator and can be cooled to 60
mK. Atoms are produced in a cryogenic discharge, thermalize through collisions with
each other and with the liquid 4 He coated cell walls, and settle into the minimum of
the trapping magnetic field. (The field profile on axis is shown.)

21
 When loading the trap, the maximum trap depth is employed, Etrap/kB 0.5 K. To
load a 0.5 K deep trap, atoms must be precooled to T ~ 0.5 K or below. Laser cooling
is a common method for cooling atoms, but it not feasible for hydrogen because an
adequate light source does not exist for driving the 1S-2P fundamental transition at
121.6 nml. Instead, the atoms are precooled through thermalizing collisions with a
250 mK liquid 'He coated surface. A 3 He- 4 He dilution refrigerator is used to cool the
apparatus. The binding energy of hydrogen on liquid 'He (EB/kB ~ 1 K) [37] is low
enough to allow the existence of an appreciable gas phase in thermal contact with
the cold surface. This method for loading paramagnetic atoms into a trap appears to
be limited to hydrogen because other atoms have too high a binding energy on liquid
'He or any other surface one could imagine.

2.2.2 Forming Atomic Hydrogen in a Radio Frequency Dis-

charge

To form atoms, a cryogenic radio frequency (RF) discharge dissociates molecules

which were initially loaded through a small capillary and frozen on the walls in the
discharge region. The discharge is located at the top of the trapping volume, in the
highest magnetic field in the apparatus (4 T) as shown in Fig. 2-2. The discharge is
simple in design (Fig. 2-3) and essentially unchanged from the first such discharges
studied [38, 39] and the discharges which have been used in the MIT hydrogen ex-
periment for the past 10 years [40]. It is a A/4 coaxial resonator with a helical
inner conductor, designed [41] to resonate at 300 MHz. The coil is tapped so as to
impedance match at low temperature (< 2 K) the 50 Q coaxial cable which carries
the RF. This allows good coupling of the RF power into the dissociator (> 90%).
Care was taken to thermally anchor all discharge surfaces and avoid the formation of
hot, liquid 4He bare regions during loading which could serve as sights of enhanced
recombination.

'The Amsterdam hydrogen trapping group had some success with a hybrid of laser cooling and
magnetic trapping [36].

22
 f=503.5 mm fused silica lens
10
copper discharge body-
copper helical coil - 44--

copper cell top - ri

silver sinter coated copper -

G-10 plastic=
liquid helium jacket-
trapping volume -

Figure 2-3: Diagram of the discharge and cell top. The discharge body and helical
inner conductor form a A/4 coaxial resonator for 300 MHz. When resonantly driven,
the radio frequency field sparks a discharge which dissociates molecular hydrogen and
forms atoms which stream into the trapping volume. The inner surface of the cell is
coated with liquid 4 He to reduce the binding energy of hydrogen atoms to the walls.
The walls of the trapping volume are made of G-10 plastic and a cylindrical space
is filled with liquid 4 He to provide thermal conductivity. Silver sinter coated copper
sheets provide a large surface area for heat transport across the 4He-copper boundary.

When loading atoms into the trap, the discharge and cell are maintained at about
250 mK and the discharge is fired for about 30 seconds in pulsed mode, with 50 Hz
repetition rate and 100 ps pulses with peak power of 10-30 W. The hydrogen flux is
at least 1013 s-1.

2.2.3 Spin Polarization

Atoms are produced in all four IS hyperfine states and carried into the trapping
cell in a puff of gaseous helium. After dissipating energy through collisions with the
walls, high field seeking H4 atoms are pulled back to the discharge by the magnetic

23
field gradient. HT atoms remain in the trapping volume and settle into the minimum
of the magnetic field through atom-atom collisions. Atoms in the c state are lost
from the trap very rapidly due to spin-exchange collisions which change the atoms to
untrapped H4 [42]. This creates a doubly spin-polarized sample [43, 44] of typically
a few times 1014 d state atoms.

2.2.4 Loss Processes and Gas-Surface Equilibrium

Two loss processes on the surface and one in the gas are important in the loading
of d-state atoms into the trap. Magnetic impurities on the wall can flip the electron
spin of a surface d state atom, causing it to be quickly expelled from the trapping
region [45]. Also, in thermal equilibrium there is a small surface density of H4 atoms
on the wall in the trapping region. While on the surface, a d state atom can readily
recombine with a H4 atom since the interaction is through the singlet potential and
the wall serves to conserve energy and momentum [46].
The hydrogen sample and wall are well thermally connected at the loading tem-
perature of 250 mK because the H_4 He binding energy is low enough that atoms
spend a short time on the wall compared to the characteristic recombination time
with residual H4 or the spin flip time due to magnetic impurities. Atoms can thus
hit the wall, stick, exchange energy with the wall, and return to the trap. The con-
stant flux of energetic HT atoms over the magnetic barrier to the wall, and off the
wall into the trapped sample, maintains thermal equilibrium.
In the gas, weak dipole-dipole collisions[42, 44] can flip the spin of a d state atom,
causing it to be lost from the trap. This process obeys the local two-body equation

hdip = -gn 2, (2.1)

where n is the local density. The loss rate constant, g = 2

(Gdd-ac +Gdd-aa +Gdd-_ad)

is the sum of the dominant decay event rates for collisions between two d state atoms
in the doubly spin-polarized sample. The event rates have been calculated[42] as a
function of the magnetic field and temperature. For temperatures below 500 pK and

24
magnetic fields below 10-2 T, the rate constants change by less than a few percent
from the zero temperature and field values. For this experiment, when the trap depth
is less than half the hyperfine energy liberated in the inelastic processes (68 mK),
atoms ending in the c and d states after a spin flip are lost from the trap, which
explains the factor of two in the expression for g. For B = 0 and T = 0, the
theoretical rate constant is g = 1.1 x 10-1 cm 3 /s. At n = 1013 cm- 3 , this implies a

100 second sample lifetime. The theory quotes no uncertainties.

There is an experimental measurement of g [44] which agrees with theory with a
17% uncertainty. It is important to note that in that experiment [44], there was no
way for d or c atoms to escape the trap besides flipping their electron spin. Hence
the d and c atoms in the exit channels of spin flip collisions remained trapped and
were not counted in the loss rate. The expression for g in [44] differs accordingly.
Since dipolar decay preferentially removes atoms from the low energy, high density
region at the bottom of the trap, it is a heating mechanism for the sample.
A final loss process worth noting is three-body recombination in the gas. When
three hydrogen atoms collide, two can recombine while the third serves to conserve
energy and momentum. This is governed by the rate equation

p3-body- -Ln 3 (2.2)

The decay constant is L ~ 10-38 cm 6/s [33]. The process is negligible in the MIT
hydrogen experiment, but it is often the density-limiting effect in alkali metal trapping
experiments. [47].

2.3 Evaporative Cooling

After the trap is loaded the wall temperature is quickly lowered, and when it reaches
about 150 mK, the wall residence time becomes so long that atoms which reach the
wall and stick[48] are lost due to spin flip or recombination. The sample thermally
disconnects from the wall and energetic atoms are quickly removed from the sample as

25
4
CD

3 -

1 -Evaporation

axial displacement

Dipolar Decay
HT+HT -> HT+H
Figure 2-4: Schematic of the magnetic trap and hydrogen cloud shortly after loading
the trap. Energetic atoms escape over a saddlepoint in the magnetic field, evapo-
ratively cooling the sample. Dipolar decay removes atoms preferentially from the
highest density region at the bottom of the trap, heating the sample. Heating and
cooling balance when the sample temperature is about one thirteenth of the trap
depth. By lowering the confining magnetic field, the sample can be further cooled
through forced evaporation.

they evaporate over the magnetic barrier (Fig. 2-4). The temperature of the remaining
sample drops until a balance is reached between the cooling due to evaporation and
heating due to dipolar decay. A useful quantity to define is q = et,,p/kBT, the ratio
of trap depth to equilibrium sample temperature, which, to a good approximation,
depends only on the trap depth[24]. In the 0.5 K deep trap, q _ 13. After the
sample thermally disconnects from the wall and cools, there are about 1014 atoms
at T = 40 mK, and the density distribution is given by n(r) = noexp(-U(r)/kBT),
where no ~ 10" cm- 3 is the peak density found at the magnetic field minimum.

26
 1000

NNN
100 - --

IN

10 -- 0mK-

10 mK-.-
. .. 300 uK
0.01 -100
30 uK

0.001'
110 10 1 10 1210 1310 1 10 15
Density [cm~3]

Figure 2-5: Time between collisions, T =_1/no-Od2 for various hydrogen densities
and temperatures. The collision cross section is or- =87a2 = 1.06 x 10-15 CM2 and
ic' = 16~kBT/7rm is the thermal average relative velocity between two atoms. The
thermalization time for a sample out of equilibrium has been estimated to be a few
collision times[49, 47].

2.4 Forced Evaporative Cooling

H. Hess, while working on the MIT hydrogen experiment, first suggested that because
the sample equilibrates at a fraction of the trap depth, one could force the evaporation
and further cool the sample by gradually lowering the confinement barrier [34]. If the

trap depth is lowered slowly enough for the sample to remain in equilibrium through
elastic collisions (Fig. 2-5), but fast enough so that dipolar decay does not remove too

many of the atoms, no increases as well [47]. The peak phase space density increases
even though atoms are lost from the trap.
The cooling power of evaporation is proportional to the rate at which energetic
atoms are created in binary collisions and removed from the sample. For ideal evap-

oration, all atoms which attain energy greater than the trap depth leave immediately

27
and the cooling power per atom varies as a constant times the collision rate per atom,
no-l' 2 oc /T<,where - is the elastic collision cross section.
Since the dipolar heating rate goes as the loss rate per atom, gn, and is indepen-
dent of temperature, the ratio of cooling power to heating decreases with temperature.
Thus, r decreases with temperature as well. Evaporative cooling of hydrogen becomes
difficult below about 1 pK [47, 50]. It was predicted that BEC in hydrogen could be
realized at about 30 pK [34] where r/ ~ 7.

2.4.1 Magnetic Field Saddlepoint Evaporation

Forced evaporative cooling was first realized in the current experiment [51] by lowering
a magnetic field saddlepoint which forms the lower axial confinement barrier and
defines strap, as shown in Fig. 2-4. If the barrier field is lowered from the starting
height of about 0.8 T, to a final height of 15-20 x 10-4 T in about 5 minutes, the peak
density increases to nearly 1014 cm- 3 while the temperature drops to about 120 1 K
[52]. Over 10" atoms remain in the trap.
Further evaporative phase space compression proves to be impossible in the current
apparatus using this method to define the trap depth and remove energetic atoms.
This was explained by the Amsterdam hydrogen trapping group [53] in the context
of a similar experiment. Evaporating over the magnetic saddlepoint only removes
atoms with high energy in the axial degree of freedom. This is not a problem when
the density of atoms is relatively low and the temperature is high because all atoms
which acquire energy greater than trap still escape from the trap and evaporation is
essentially ideal. This happens because the low density implies that the mean free
path between collisions is long, and at a high temperature, atoms access anharmonic
regions of the trap. Energy is exchanged between the three motional degrees of
freedom quickly and the time required for atoms to explore the entire trap and escape
is short compared to the time between collisions.
However, when the density increases, the mean free path decreases, and when the
sample temperature drops, atoms settle into the harmonic region of the trap and the
coupling between the degrees of freedom decreases. It becomes more likely for atoms

28
 25 I
ITt>
1S, F=1 d .

E c
-2
-0 l
b 4.
40
a>,
20 0
HT
LU
IU I-

-200

5 -400
0 0.1 0.2 0.3 0.4 0.5

0 0.002 0.004 0.006 0.008 0.010

Magnetic Field [Tesla]
Figure 2-6: Zeeman diagram of iS F =1 states in low magnetic fields showing the
transitions driven for RF evaporation. A trapped atom is in the d state. When it
climbs the magnetic potential and passes a region where the field brings its levels into
resonance with the RF frequency, the electronic spin can flip, changing the atom to
a c state atom which feels a much weaker confining potential. If the c state atom
does not escape the trap, in ensuing passes through resonance, it can be changed to
an anti-trapped b state atom and be expelled from the trap. The inset shows the
complete IS Zeeman diagram.

with radial energy above sEtrap to suffer a collision before the energy is transferred into
the axial degree of freedom. The additional collision is most likely to redistribute
the energy so that neither the energetic atom, nor the atom with which it collides,
has enough energy to escape. The probability for an energetic atom to escape drops
significantly and the evaporation becomes nonideal, increasingly "one-dimensional,"
and inefficient [47].

29
2.4.2 Radio-Frequency Evaporation

To overcome this problem, an alternative method of defining the trap depth and
removing energetic atoms is used. A d state atom's spin can be flipped by an RF field
which is resonant with the magnetic sublevel spacing as shown in Fig. 2-6. (This is
reminiscent of electron spin resonance spectroscopy.) In an inhomogeneous magnetic
field such as the trap, the RF frequency, P, defines a resonance surface where the
magnetic field has constant magnitude, B ~ hv/pB. Atoms with enough energy in
any motional degree of freedom can climb the magnetic potential and pass this surface
and be ejected from the trap. Lowering the RF frequency forces the evaporation. This
method for ejecting atoms from a magnetic trap was introduced by Pritchard et al.
[54] and has been used to evaporatively cool Rb [19], Na[20] and Li[21, 22] atoms into
the quantum degenerate regime.
In this experiment, the RF evaporation typically starts at a trap depth of 1.1
mK, corresponding to a frequency of 23 MHz and sample temperature of 120 /1 K. In
about 25 seconds the trap depth can be lowered to 100 pK, producing atoms with a
temperature below 30 [pK.
Implementing RF evaporation in the current experiment required major modifi-
cations of the trapping apparatus to reduce RF eddy current heating[23]. Virtually
all metal in the trapping region had to be removed because more than ~ 100 PW of
RF heating in resistive conductors would cause the cell temperature to rise enough
to create a significant vapor pressure of He in the cell which would greatly diminish
the sample lifetime. Thermal transport along the trapping cell, which was previously
supplied by copper wires, is now provided by a jacket of liquid 4 He (Fig. 2-3).

2.5 Probing the Trapped Gas

Many common methods for alkali atom detection, such as hot wire ionization for ex-
ample, do not work for hydrogen because of its large electron binding energy. When
hydrogen atoms leave the trap, however, they can recombine on the walls, and in a
cryogenic environment, the liberated recombination energy can be recorded on a sensi-

30
 I . 1 I I I

, 300
2-. Temperature = 145 IK
e250 Atom Number = 2.5 x 1011
I-2
- Peak Density - 2.6 x 101 cm
E 200 -
0A
-5\ %
0 150 -\ \ Bolometer Signal
--- 135pK
\ -145 K
a100 - -- 55 A

50
0

-0-
U) .. .

' - I I I I

0 0.5 1.0 1.5 2.0 2.5

Barrier Height [mK]

Figure 2-7: Bolometric determination of the sample temperature. The power de-
posited on the bolometer is recorded as a trap confinement barrier is lowered. The
power measures the number of atoms with energy equal to the barrier height. The
smooth curves are the expected distributions for various sample temperatures.

tive bolometer[24, 55]. A robust and reliable method of measuring the temperature[56]
and density[51] of the hydrogen sample is to record the recombination energy while
lowering the magnetic barrier and releasing the atoms from the trap.

2.5.1 Bolometric Temperature Measurement

If the barrier is lowered slowly compared to the time for atoms to escape, recombine,
and deposit energy on the bolometer, the power deposited is proportional to the
number of atoms in the sample at the energy of the barrier. Provided the release
time is shorter than the collisional rethermalization time, the thermal distribution of
atoms in the trap will not change during the dump, and from the bolometric data and
a knowledge of the magnetic field, one can find the sample temperature (Fig. 2-7).
Considering the time scale constraints mentioned above and experimental de-
tails such as signal to noise of the measurement and inductive time constants of the
magnets, the bolometric determination of temperature is only reliable for samples

31
 4.5 . ,
4.0
3.5
2 3.0
2.5
2.0
1.5
1.0
0 50 100 150 200
Time [seconds]

Figure 2-8: Decay of a trapped hydrogen sample. The sample density is found from
the slope of N(0)/N(t), the inverse of the normalized total number of atoms remaining
in the trap. The data shown indicates a density of 6.0 x 1013 cm-3.

temperatures between 100 pK and 5 mK and densities below 1014 cm-3.

2.5.2 Bolometric Density Measurement

One can measure the sample density by releasing the atoms from the trap after holding
them for different times following the forced evaporation. The total recombination
energy deposited on the bolometer by a sample held for a time t is proportional to
N(t), the number of atoms remaining in the trap at time t. N(t) decreases with time
due to dipolar decay and evaporation, as shown in Fig. 2-8.
By integrating the local dipolar decay loss rate (Eq. 2.1) over the trap volume,
one finds
V2(T)
Ndip = -g 2 (;) N 22 (t), (2.3)

where Ngdi is the atom loss rate due to dipolar decay, and Vm(T) f d3 , e-mU(r)/kBT
is an effective volume.
While measuring the decay, atoms are also lost due to evaporation. To maintain
thermal equilibrium, for every r7 - 2 atoms lost due to dipolar decay, 1 atom must

32
evaporate[51]. Thus N = Ndip + Nevap - NAdip( - 1)/(7 - 2), and

N(0)
N(t)
1 (T - I)
= 1 +
( - 2
V2 (T)
V1(T)
gno(0)t(24 (2.4)

where no(t) is the peak density in the trap at time t, The value of V2 (T)/V 1 (T)
is found numerically for each trap configuration, but to a good approximation, one
can describe the potential by the potential energy density of states exponent, 6:
d'r oc U6- 1 dU. In a linear quadrupole trap, 6 = 2 and V2 (T)/V 1 (T) . (1/2)6 ~ .25.
For cold samples, which have settled nearer the bottom of the trap, the potential is

harmonic in the radial direction, not linear, and 6 can be significantly different.
From the relative number of atoms measured with the bolometer, and a knowledge

of the trap shape, one can extract the initial peak density. The number of atoms in

the sample and the absolute sensitivity of the bolometer can then be found from

N(0) = V (T)rno (0).

The slope of the sample decay curve is reproducible to a few percent for identi-

cally prepared samples. The dominant uncertainty in this measurement, however, is

systematic, arising from imperfect knowledge of our trapping fields. This limits the

accuracy of the measurement of no(0) to about 10-20%. Any error in the calculation

of g would also be reflected in the inferred densities.

At temperatures below about 100 pK the trapping magnetic fields are so low that

additional contributions from magnetic materials and trapped fluxes in the supercon-

ducting magnets become significant and the trap is not well known. The bolometric

method for measuring no then becomes unreliable.

Probing the trapped gas with bolometric techniques is limited by the escape time
of the atoms and it also necessitates the destruction of the sample. Spectroscopic

methods offer the possibility of monitoring the gas in situ. The Amsterdam hydrogen
trapping group implemented Lyman-alpha 1S-2P spectroscopy of the sample [57],

and then, to overcome the limitations imposed by the large natural linewidth of
the 1S-2P transition, they developed resonance enhanced two-photon spectroscopy

(RETS). In RETS, two-photon excitation to a relatively long-lived 3S or 3D state

33
is enhanced by tuning one laser frequency near resonance with an intermediate 2P
state [58]. The MIT hydrogen trapping group chose an alternative path and pursued
two-photon 1S-2S spectroscopy as a probe of the trapped gas.

34
Chapter 3

Overview of 1S-2S Spectroscopy in

a Trap

This chapter gives an overview of the 1S-2S spectroscopy component of the experi-
ment, as well as a bit of historical perspective. The laser system was designed and
built by J. Sandberg and C. Cesar, and more details are available in their Ph.D.
theses [59, 60].
High resolution spectroscopy is a useful probe of the trapped hydrogen gas, but the
observations reported here also show that the physics of the excitation is of interest
by itself. The two-photon spectrum is novel, and the long coherence time of the
laser-atom interaction makes the atomic motion much more important than it is for
normal one-photon transitions. The excitation also probes atom-atom interactions
and correlations in ways which are not yet completely understood.

3.1 Background

The two-photon 1S-2S transition was first observed in 1975 in a gas cell[61], and the
experimental linewidth was limited to about 100 MHz by the pulsed laser source.
Improvements in nonlinear optical frequency generation made CW experiments pos-
sible, first in a gas cell at about 0.2 torr [62] where the linewidth was collisionally
limited to a few MHz, and then in a liquid nitrogen temperature atomic beam [63]

35
where the linewidth of 50 kHz was due to the second order Doppler-shift and finite
interaction time of the atoms with the laser. By cooling the atomic beam to about
5 K, and selecting the signal from only the coldest atoms, the resolution has reached
about 2 kHz[64].
By comparing the 1S-2S transition frequency to that of another electronic hy-
drogen transition, one can determine the Rydberg constant and Lamb shift [8, 9],
and the deuteron radius can be determined by comparing the 1S-2S frequency in
hydrogen and deuterium [8, 65]. The 1S-2S frequency in hydrogen is fis-2s =

2, 466, 061, 413,187.34(84) kHz [9], and this is the most accurately known frequency
in the UV or optical region.
In the late 1980's the MIT Ht trapping group, in the quest for BEC, set out to
excite the 1S-2S transition in a trapped sample in order to study the gas in situ.
A secondary goal was high resolution spectroscopy of the cold atoms. In a trap the
possible interaction time is long, and at the low thermal velocities, the second order
Doppler-shift is negligible.

3.2 Two-Photon Excitation

Figure 3-1 shows a sketch of the levels involved in 1S-2S spectroscopy of trapped
hydrogen. The angular momentum is zero for both the 1S and 2S states, so the
transition cannot be driven by one photon. An atom can absorb two 243 nm photons,
however, and be excited to the 2S state through an intermediate virtual P level. For
a transition to occur, an atom must absorb two photons in a time less than h/A,
where A = (Ep - hv) is the laser detuning from Ep, the energy of the P level. The
excitation rate varies as the square of the laser intensity, and for a given intensity,
the rate is much lower than for a one photon transition. The 2S state is metastable
(Tis = 122 ms) because an unperturbed 2S atom can only radiatively decay to ground
through emission of two photons[66].

36
 25 122ms --...... Stark mixing by E-field
-----------------------
2P1.6ns

C'4

E Lyman,-photon 121.6nm

CB

55 decay into all sublevels

Figure 3-1: Level scheme for 15-2S spectroscopy of magnetically trapped atomice
hydrogen. Trapped F = 1, mF = 1, 1S atoms are excited to the metastable F =
1, mF = 1, 2S state by absorption of two 243 nm photons. An applied electric
field Stark mixes some 2P character into the excited state wavefunction and causes
prompt radiative decay to the ground state through emission of a single Lyman-alpha
photon (121.6 nm). 1S and 2S F = 1, mF = 1 atoms see the same magnetic trapping
potential.

3.3 Detection Scheme

The sample is optically thin to the laser radiation, even on resonance, so it is not

feasible to monitor the excitation rate by measuring direct absorption of the laser.
However, photoexcitation can be detected by monitoring fluorescence from the excited
state. The signal to noise ratio can be greatly enhanced by using a pulsed scheme (Fig.
3-2), in which each excitation pulse is followed by the application of a short electric
field pulse (-- 10 V/cm). This Stark mixes some 2P character into the excited state

wave function (See Sec. 4.4.3), and causes 2S atoms to promptly decay to the ground
state through emission of a single Lyman-alpha photon (121.6 nm) which can be
detected by a microchannel plate (MCP) [67, 68].
A typical timing sequence for excitation and detection is shown in Fig. 3-3. A

37
 M
f=503.5 mm r5 0 cm
fused silica lens
0
00
48 cm
243 nm, laser

cm
46

electric field wires 2 cm

C5w/

trapped atoms o cm

V.-

Lyman-alpha photons -2 cm
-4

R=250 mm
BK7 mirror

MgF 2 window -24 cm

MgF2 Lyman-alpha filter

-30 cm

24 Mn 60 V
1155"VV16 pF
MCP assembly - - --- - 50 k( ) | -32 cm
2100 V to amp ler

-34 cm

Figure 3-2: Excitation and detection. Trapped atoms are excited to the 2S state
by the 243 nm standing wave laser field which passes on axis. The laser is blocked,
and ±100 V is applied across the electric field wires, producing a 10 V/cm field in
the cell. This causes the 2S atoms to rapidly decay through emission of 121.6 nm
Lyman-alpha photons. Approximately 10- of these photons are detected with a
microchannel plate assembly. The aspect ratio of the figure is 1:1

38
 I I I I

background counter

signal counter

applied E field

stray E field counter

laser counter

laser

-500 0 500 1000 1500

Time [pus]

Figure 3-3: Typical timing sequence for 1S-2S excitation and detection.

mechanical chopper modulates the laser beam with 50% duty cycle and 400 Ps pulse
length. The electric field pulse is typically 12 ps long, and the resulting signal pulses
are recorded by a counter enabled during this time. During this Stark quench the
peak count rate can exceed 10 MHz. The number of signal Lyman-alpha counts
recorded as a function of laser frequency is the photoexcitation spectrum.
Additional counters during each timing sequence are used for diagnostic purposes.
The laser power can vary 10-20% during the recording of one spectrum, and its level
is monitored by a counter gated open for 25 ps during the laser pulse. The count
rate from scattered laser photons, which can exceed 100 kHz, is proportional to laser
power. Following the laser pulse, but before the Stark quench, a counter enabled
for 25 pus monitors the Lyman-alpha fluorescence caused by stray electric fields. The
count rate can be used to measure the value of the field. Typical stray fields in the
cell are on the order of a half volt per centimeter or less, and give the 2S state a
lifetime of milliseconds (Sec. 4.4.3). A counter after the quench records dark counts
(- 200 Hz) for 25 /is. These counts arise chiefly from long lived fluorescence from
organic materials in the cell which were excited by the laser, and the measured rate
can be used to establish the background count rate, which can be subtracted when
measuring very weak 1S-2S signals.

39
 This sequence is repeated for typically 10-100 laser pulses for each laser frequency.

3.4 Microchannel Plate Photon Counter

The MCP assembly is shown in Fig. 3-2. The top plate is a 50 mm diameter, 0.6 mm
thick lead glass disk with 60% open area in the form of an array of 10 pm diameter
channels[68]. Each channel has a length to diameter ratio of 60:1. The top surface
is coated with CsI to decrease the work function for efficient production of electrons
by impinging vacuum ultraviolet photons. The top surface is positively biased by 60
volts with respect to the surrounding housing to guide electrons into the channels.
Approximately 1000 volts is applied across the channels to accelerate the electrons.
Collisions with the channel walls eject more electrons and create an electron shower.
Each channel acts as an electron multiplier with a gain of about 104 .
The bottom plate is similar in structure to the top, but it has a length to di-
ameter ratio of 40:11 for its channels and is uncoated. With 2 plates, the gain is
about 106 so that a single photon results in a 1 mV, 5 ns pulse into 50 Q which is
capacitively coupled to a high bandwidth video amplifier (x100). Amplified pulses
are discriminated and turned into logic pulses which are counted at rates up to 100
MHz. The quantum efficiency of the assembly for 121 nm photons was calibrated
against a Hamamatsu R972 photomultiplier tube [69] and found to be 25%. Due to
a small solid angle (~ 10- sr), absorption of Lyman-alpha in optical elements, and
MCP quantum efficiency, only 10-' of the emitted photons are detected.
To be close to the atoms, the MCP is mounted inside the cryostat. At low tem-
perature, the replenishment of the charge in a single channel after it fires can take
seconds. During the recovery time, that channel is effectively blind to photons. When
a significant fraction of the channels fire during one recharging time, the quantum ef-
ficiency of the MCP drops. This implies a maximum sustainable counting rate which
is about 20 kHz at 20 K, 200 kHz at 80 K, and greater than a MHz at room tem-
perature. If precautions were not taken, counts due to scatter from the laser would

160:1 plates are newly available and are superior in gain and mechanical strength.

40
 Coherent 699
Ring Dye Laser SHG ring cavity Doppler

Kr+ Laser
PEOM BBO
30 meters
innova 200 1
486 nm 243 nm
Frequency
control PZT
N AOM

Ref Cavity A-) Position

/ control
~+~PSD
-~

EOM mechanical chopper - -

Trapped
Tellurium Spectrometer Hydrogen.
1
"Te 2
/
AOM /
Oven Fluorescence Fluorescence

MCP

Figure 3-4: Laser system for 1S-2S spectroscopy of trapped atomic hydrogen. Com-
ponents are explained in the text.

saturate the MCP even though the sensitivity to 243 nm photons, according to the
manufacturer, is at least 4 orders of magnitude lower than to 122 nm photons. The
MCP would be left blind during the Stark quench since the recovery time is longer
than the time between laser pulses. The MCP assembly is weakly anchored to 4 K,
but it is normally heated to 70 K or higher during operation. In addition, a Lyman-
alpha filter[70] above the MCP cuts the 243 nm light by a factor of 500 and drops
the laser scatter count rate below the saturation level. Unfortunately, the filter also
cuts Lyman-alpha transmission by 90%.

3.5 Laser System

The laser system used for spectroscopy of atomic hydrogen (Fig. 3-4) was based on
a design from the group of T. Hinsch in Munich [71]. We briefly point out the

important features here. (For further details on some of the optical devices discussed
below, see [72].)
A Coherent 699 ring dye laser, pumped by a krypton ion laser, produces about 450

41
mW of linearly polarized 486 nm radiation in a single longitudinal TEMOO mode. Due
to acoustic vibrations of the laser support structure and fluctuations in the thickness
of the dye lasing medium, active stabilization is required to maintain a narrow laser
frequency spectrum. An intracavity electro-optic modulator, piezo-mounted mirror,
and galvo-mounted brewster plate are used to control the optical path length in the
laser cavity and lock the frequency to the center of a transmission fringe of a stable
Fabry-Perot optical resonator. The error signal for the frequency control electronics
is generated using the Pound-Drever-Hall locking scheme [73]. Using the hydrogen
spectrum as a reference, the resulting spectral linewidth has been shown to be below
500 Hz for times of a few seconds (See App. A). The frequency of the laser can be
tuned with respect to the fixed transmission fringe by changing the RF drive frequency
of an acousto-optic modulator (AOM) which is in the light path to the the optical
resonator.
Absolute frequency calibration to within a few hundred kHz is determined with
Doppler-free saturated absorption spectroscopy of a transition in molecular 13 0
Te2 .
Appendix E provides more details on this important component of the experiment.
The 243 nm radiation necessary to drive the 1S-2S transition is generated by
frequency doubling the 486 nm light in a beta barium borate crystal (BBO)[74]. To
increase the nonlinear efficiency, the blue power is enhanced in an optical build-up
cavity place around the crystal. The Hiinsch-Couillaud locking scheme[75] is used
to maintain the cavity on resonance with the 486 nm laser light. With 450 mW of
486 nm light from the laser, about 20 mW of linearly polarized 243 nm light can be
produced. The 243 nm beam is highly astigmatic due to the large double refraction
angle of BBO, and, even with astigmatism compensation, only about 10 mW of power
is in a useful TEMOO mode.
The laser and hydrogen trap are in separate rooms, 30 meters apart, so the 243
nm beam is collimated and sent through air to the cryostat. The pointing stability is
servo controlled using piezo-mounted mirrors, quadrant photodiodes, and a feedback
system designed by C. Cesar [60]. Optics inside the cryostat (Fig. 3-2) focus the
beam to a waist radius of 20-50 pm in the atom cloud to produce the high intensity

42
necessary to excite atoms to the 2S state. The beam is retro-reflected back on itself
by a mirror below the atoms to create a standing wave in the trap region. The return
beam is monitored on the optics table to ensure proper overlap with the outgoing
beam. Continuous manual adjustment is required to maintain the overlap. This is a
serious limitation for the experiment because it can take up to a second to acquire
proper alignment when the laser beam is first let into the cryostat.
The laser can be modulated by a mechanical chopper at frequencies up to 2 kHz.
An AOM can be inserted in the 243 nm beam path to allow faster chopping and
active control of the excitation power at the expense of a 50% reduction in power.
In these experiments, the AOM was only used when testing the limits of the spectral
resolution of the 1S-2S signal as discussed in App. A.

3.6 Photoexcitation Spectrum

The metastability of the 2S state results in the narrow natural linewidth of the IS-
2S transition (1.31 Hz at 121 nm). This makes the spectrum extremely sensitive
to potentially interesting broadening and shift mechanisms which provide valuable
information about the sample. These effects are small in the cold, dilute gas, and
they would be hidden by the natural linewidth of a one-photon transition. (The 1S-2P
single-photon transition linewidth, for example, is 100 MHz, and, as will be described
later, the maximum shifts we see are about 1 MHz.) In addition, the narrow linewidth
can be exploited for ultrahigh resolution spectroscopy and fundamental measurements
[10, 9].
The transition energy for a trapped 1S atom with momentum pis, which absorbs
two photons with momenta hki and hk 2 and goes to the 2S state, is given by

2
h1/Iaser + (inc 2 + ES 2s)2 _ P sc2 + (Mc 2 ) 2
Pis - (hki + hk 2 ) 1hki + hk 2 2
_Pis + hki + hk 2 2 EIS-2s
Eis22 + -+2 iM 2m 2m Mc2

(3.1)

43
(We have used the fact that the atom must take up the momentum of the photons
SO P2s = pis + hki + hk 2 .) The second and third terms represent the first order
Doppler-shift and the photon momentum recoil shift respectively. The last term in
Eq. 3.1 is the second order Doppler-shift. This effect often limits the resolution for
1S-2S spectroscopy in an atomic beam, but it is completely negligible for trapped
hydrogen. (At 100 pK, v 2 /c 2 ~ 10-, and the shift is 10-2 Hz.)
Since the excitation occurs in a standing wave, atoms can absorb photons in two
ways. Absorption of two co-propagating photons (ki = k2 ) gives rise to a recoil shifted
(AVrecoii = 6.7 MHz) and Doppler-broadened feature, and is called Doppler-sensitive
excitation. The Doppler-width of this feature (RMS width AVDopper = 0.374 X 109
Hz K- 1/ 2 ) is a valuable absolute measure of the temperature of the sample (Fig. 3-5).
Doppler-free excitation results from the absorption of two counter-propagating
photons (ki = -k 2 ). There is no recoil shift or Doppler-broadening in this case and
the resulting spectral feature can be very narrow. The dominant broadening for low
density samples usually comes from the finite interaction time of an atom with a laser
(time-of-flight)[10, 76, 77]. In high density samples, the resonance is red-shifted and
the feature is inhomogeneously broadened by atom-atom interactions [11]. This effect
is often called the cold collision frequency shift [15, 42, 16, 17, 12, 18]. The lineshapes
observed under different experimental conditions are discussed in Chap. 4.

44
 5000 1 1 1 1 1 1
/WN\/ 0 N\
4000 Doppler-Free Excitation

0 3000

2000

1000 Doppler-Sensitive Excitation

x10

0
-2 0 2 4 6 8 10 12
Laser Detuning [MHz at 243 nm]
Figure 3-5: Composite 1S-2S two-photon spectrum of trapped hydrogen. The intense,
narrow peak arises from Doppler-free absorption of counter-propagating photons. The
wide, low feature on the right is from Doppler-sensitive absorption of co-propagating
photons. The width of the Gaussian fit to the Doppler-broadened lineshape implies a
sample temperature of 40 pK. Zero detuning is taken for unperturbed atoms excited
Doppler-free. The Doppler-sensitive feature is shifted by 6.7 MHz due to momentum
recoil. All frequencies refer to the 243 nm excitation radiation.

45
Chapter 4

Formal Description of 1S-2S

Two-Photon Spectroscopy

In order to interpret the observed 1S-2S spectra, we present the formalism describing
two-photon Doppler-free and Doppler-sensitive excitation. An effective Hamiltonian
is constructed with which one can derive lineshapes and transition rates using for-
malism that is familiar from one-photon transition calculations. Perturbations such
as the cold collision frequency shift can be added to the Hamiltonian. Experimental
data is presented.

4.1 1S-2S Two-Photon Transition Theory

4.1.1 Physical System and Interaction

The ground state for the transition is the F = 1, mF= 1, IS state, and the excited
state is the F = 1, mF = 1, 2S state (Fig. 3-1). The transition must proceed through
intermediate P states. The two-photon 15-2S selection rules preclude any change in
hyperfine state during the excitation - the electronic and nuclear spins are essentially
along for the ride. For our purposes, the analysis must incorporate atomic motion.
The translational state wave functions are described by plane-wave states with peri-
odic boundary conditions and normalization in a box of volume V. The IS and 2S

46
states have momentum pi and P2 respectively, and the intermediate nP state has
momentum p' .
The important atomic energies are

Es,p PI2 (4.1)

2m
P2 2
E 2 s,P 2 = hWis- 2s + 2
2m
/2
EnPp, = hWlS-nP + Pn-
2m

We define

hWIS,pi -2S,P2 = (E2s,P2 - Eis,pl) = h01s-2s + P .22 (4.2)

2m 2m

The first order Doppler shift of the transition frequency will arise naturally from
momentum conservation. This treatment neglects the second order Doppler shift, but
this effect could be recovered by considering energy and momentum relativistically.
The atoms actually move in a trap and do not simply occupy plane wave states.
Our treatment will preserve all the relevant effects of the atomic motion so long as
the trap level spacing is small compared with the motional energy of the atoms. This
condition is not fulfilled for atoms in a Bose-condensate, and the discussion of the
1S-2S spectrum of BEC is deferred to Chap. 7. Cesar [60] developed a two-photon
transition formalism in which the atomic motion in the trap is quantized, and it is a
useful alternative picture.

4.1.2 Excitation Hamiltonian

The derivation of the photo-excitation spectrum begins with time-dependent pertur-

bation theory[78], with the perturbing Hamiltonian

H' = -ei -E(R, i, t). (4.3)

47
The operators are R, the center of mass of the atom, and i, the position of the
electron with respect to the nucleus1 . The charge of the electron is e < 0, and we
represent the standing wave laser as a classical monochromatic electric field of the
the form

E(R, f, t) = !@1E1(R)eiki(+r± -iwit + 2 E 2 (R)eik2(f)-iw2t

F2 + c.c. (4.4)
2 2

E1 and E 2 are taken to be real and they contain the slow spatial variations of the
beam profiles. The laser is applied at time to.
The intensity of beam i is 1i = 1EocjEj(R)j 2 . For the experimental situation, we
have approximately a TEMOO Gaussian mode,

I(R) = P exp X +Y2 . (4.5)

2P
TW2(z) ex [ 2(92 + y2 )-1
w2(z)

Here, P is the power in the laser beam (4-8 mW), and w(z) = wo 1 + z 2 /z2 is the
beam radius at position z. wo is the beam waist (20-50 pm), and zo = 7rw/A is the
divergence length or Rayleigh length (0.5-3 cm). A typical peak intensity in the laser
focus is about 200 W/cm2
As is well known, (2S, P2 I H'(t") IIS, pi) = 0, because parity conservation forbids
a one-photon Al = 0 transition, so the first order transition probability vanishes. We
find the transition rate from the second order coefficient of the 2S state,

(t) = - ~jdt" dt' 1 (2S, P2 H'(t") I nP,p')e-wn-2s,2(t -o)

h I ft~o n,pn.

(nP,p' I H'(t') I 1S, pi)e-si1S,pji-nP,pn(tf -to). (4.6)

The sum extends over all P states, including the continuum. By making appropriate
assumptions, we can recast this expression in the form of a first order transition

'To be explicit, when a position variable is an operator, it will have a tilde.

48
coefficient,

C sf (t) - dt"V2s,P2 ;1s,P1 (tII)eiW1,p1 -2S,P 2 (t /-to) (47)

2
V2S,p 2 ;lS,pi (t") is the off diagonal element of the effective two-level Hamiltonian.
To calculate V2S,p 2 ;1S,pl(t), we make the rotating wave approximation and the adi-
abatic approximation [79] (jdEj(R)/dtj < jwjE(R)j). Using the electric dipole ap-
proximation, we set eiki-r to one, but leave eikiRf since it contains important informa-
tion on momentum exchange during the excitation. We specialize to a standing wave
laser field, for which k, =-k2- k and wi = W2 _ w; also, h w+wis,np I -

Appendix B gives more details of the calculation, but we present the result here,

V 2 s,P 2 ;1s,p1(t) = h ( )' 3

6
r2hC{[I(R)6 p2,P1+2hk - I2(R) p 2 ,p1 -2k]Msls/
2

+ Ii(R)I2 (R)p2 ,p1Mise 2

iwt. (4.8)

Sums over dipole matrix elements, r,, have been reduced to

MS, is =
(2R&o 3 37 3 e 2
(r2s,nP * ci rpiS - +- r2S,nP - Fj rnPJS - Ej).
ak F2h 2n (w + WS-nP)
(4.9)

The term proportional to M 1 in Eq. 4.8 gives rise to Doppler-free absorption of

two counter-propagating photons, for which there is no momentum transfer and thus
no recoil shift. The term proportional to M2lIs gives rise to recoil-shifted Doppler-
sensitive absorption of two co-propagating photons, for which the photons can come
from either of the two laser beams. In our particular experimental arrangement, the
polarization is linear and = 2, w ws2s/2, and Mj = s = 11.78 [80].

Using Eq. 4.8 alone one can calculate the 1S-2S spectrum in the absence of any

2V is called the "effective two-level Hamiltonian" because only reference to the initial and final
levels remains.

49
additional perturbations. Experimentally, this is often a good approximation, so we
discuss this situation before including additional spectral shifts and broadenings.

4.2 Doppler-Sensitive Excitation

Theoretical Description

Doppler-sensitive excitation results in a spectral feature with a linewidth much larger

than the inverse of the time of excitation. It is thus valid to define the Doppler-
sensitive excitation rate for a single atom through Fermi's Golden Rule [78],

Wis,p 1j- 2 s(2hw) =J: Ws,pj-2s,P2 = 2V2S,P ;IS,pi1 2 6(E

2 2s,P 2 - Eis,p, - 2hw).
P2 P2
(4.10)
We substitute the Doppler-sensitive excitation terms of Eq. 4.8 for V 2 SP2;I S1 1. The
interaction vanishes unless the excited state has P2 p, ± 2hk, so the sum over final
P
momentum states is trivial. It yields the Doppler-sensitive excitation rate to the 2S
level

WIs,p 1- 2 s(2hw) =
2
[(R)6
1Mm
2hw - hwisS- 2 s - 2h 2 k 2
m
- 2hk.

+ Q (R)6 2hw - hwis- 2s - 2 m2k22hk -pi

m in)]
(4.11)

We have defined the Rabi frequency for Doppler-sensitive excitation by a single laser
beam,

= 21V2S,pl+2hk;1S,p1
-(R) =vi 2,3 1 = 4.632 I(R) s3 cm 2 W-1.
- h2S,1 k2R} 37r2hC~i2 \) 462 ()
(4.12)
Note that because the final states (P2 = pi ± 2hk) are not the same, we must add
the probabilities of being excited by I, and I2. We do not coherently add the 2S
amplitudes arising from each process.
When the transition is resonantly driven by the Doppler-sensitive term in Eq. 4.8,

50
the laser frequency must satisfy

2hk -Pi 2h 2 k2
2hw = E2sP2- Eis,p, = hwls-2s i + . (4.13)
m m

The recoil shift is given by the term which is quadratic in k. The k -pi term is what
is normally added in an ad hoc fashion as the Doppler-shift. The usual description of
a photo-excitation spectrum does not mention the momentum recoil, and thus cannot
arrive at Doppler-broadening in this way. The Doppler-broadening arises naturally
if momentum transfer is correctly considered as is done here. Because of the small
hydrogen mass, large transition frequency, and low sample temperature, the 1S-2S
transition in trapped hydrogen is unique in that the magnitude of the recoil shift is
greater than the Doppler width.
The experimentally observed excitation rate at a given frequency is found by
integrating Eq. 4.11 over the sample density distribution and momentum distribution.
We define the z axis along the laser beams. Note that Eq. 4.11 is independent of px and
py and the spectrum maps the axial momentum distribution. Assuming a Maxwell-
Boltzmann distribution of Pz, after a little algebra, one arrives at the Doppler-free
spectrum,
-
2
l 1/ 2 1 nc2 V _ "1S-2S _k2

1 ( m c2
exp [_2 2kB
S(v) =
/ 2 kBT V2 2 kBT V

x d3R n(R)[Q2(R) + Q2(R)]. (4.14)

This gives the number of 2S excitations per second and the frequency has been ex-
pressed in Hz. The spectrum is centered at

V1S2S hk 2 _
1
S-2S(.5
v i- + -k _ is2 + 6.70 MHz, (4.15)
2 27rm 2

and has an RMS width of ABT= s2 0.374 x 109V/T Hz K-1/2. The full

width at half maximum is 2 -21n(1/2)Av. The integral over position in Eq. 4.14
accounts for the geometric overlap of the atom cloud and laser beams.

51
 -150

0 S125 -

-J

)100--

C
0

0~4
1 50 -

2 4 6 8 10 12
Laser Detuning [MHz at 243 nm]

Figure 4-1: Doppler-sensitive excitation spectrum of a sample held at a trap depth

of 280 pK. Zero detuning is the unperturbed 1S-2S frequency. The smooth curve
is a Gaussian fit to the data which gives a RMS width of 2.3 t 0.4 MHz, indicating
T =_37 ± 12 pK.

Experimental Observation

The Doppler sensitive spectrum is an important diagnostic of the temperature of the

trapped sample. An example is shown in Fig. 4-1 for a 37 pK sample.

A calculation of the expected peak excitation rate is instructive. The laser power
was about 5 mW and the beam focus was about 45 pm, which implies a peak laser
intensity of 150 mW/cm 2 and a Rabi frequency of 700 s-1. On resonance this gives

a peak excitation rate of 0.17 s-1 per atom in the laser. The cloud is larger than the
laser excitation region, however, and numerical simulations predict that one effectively
accesses one sixth of the atoms. This implies that the lifetime of the sample due to
laser excitation is about 35 s. This is long compared to the observed ~ 10 s lifetime
when the laser illuminates the gas. The lifetime seems to be reduced by collisions
with helium atoms which the laser boils off the retroreflecting mirror.

52
 9 atoms in this sample. Assuming the laser was well
There were about 5 x 10
aligned with the atom cloud, the observed count rate implies a detection efficiency
for the fluorescence photons of about a = 10-6. This is reasonably consistent with
our estimate of a =10-5 (Sec. 3.4) considering the crudeness of all the assumptions.

4.3 Doppler-Free Excitation: Simple Lineshapes

For Doppler-free excitation, the interaction vanishes unless initial and final momenta
are equal, and there is no recoil shift or Doppler-broadening. The observed linewidth
can arise from several different processes and we discuss the lineshape resulting from
each.

4.3.1 Atoms Nearly at Rest

Theoretical Description

For an atom nearly at rest, which is in the laser beam for a long time compared to
the shortest decoherence time in the system, it is useful to use Fermi's Golden Rule
to find the excitation rate.

Wis,p,- 2 s,p1 (2hw) = 20Q(R)6(2hw - hwis-2s). (4.16)

The Rabi frequency for Doppler-free excitation in a standing wave, identified from
Eq. 4.8, is

Qo(R) = 2M2,s 2 37r 2hc Ii(R)12 (R) = 9.264 Ii(R)12 (R) s-1 cm 2 W 1 .
(4.17)
To derive the spectrum, one convolves Eq. 4.16 with the spectrum of whatever sets
the decoherence time, and then integrates the transition rate over the velocity and
momentum distribution in the sample. For example, if the natural linewidth of the
transition, -Y = 8.2 s-1, sets the decoherence time, one must convolve Eq. 4.16 with

53
the Lorentzian distribution of transition frequencies, w's2s,

G(wlss 1(ws-2-
- W-IS-2s) =y,2. W1S2s) 2 (4.18)
+ y2/4

This yields

S(w)=
= dR d 3p 1 n(R, p) Q2 (R) dwls- 2 sG(wfs s - is2s)6(2w W cs-2s)
f2 0 j 2 -y2rIS2

= d3R d3pi n(R, p) Q2 (R) _(/21r

J 2 (WIS-2S - 2w) 2 + y2 /4

On resonance, for N atoms in the laser focus,

S (WsO-2S) N 8 5 8 I1I2 -1 cm 2
W- 1 . (4.20)

As pointed out by Sandberg [59], for I = 0.89 W cm- 2, S/N = -y and the transition
is saturated for an atom in the laser beam for more than a natural lifetime.
The treatment is similar if the observed linewidth is dominated by the laser fre-
quency spectrum.

4.3.2 Atoms in Motion: Time-of-Flight Lineshape

Theoretical Description

If an atom is in the laser beam for a time which is short compared to any decoherence

time, such as the inverse of the laser linewidth, then the photoexcitation spectral

linewidth is determined by the finite interaction time of the atom with the laser, as
one would expect from the energy-time uncertainty relation AEAt ~ h. One cannot

resort to Fermi's Golden Rule; instead we compute the probability that a single atom
is excited during its pass through the laser beam. Each trajectory corresponds to

a different probability of excitation. The observed spectrum is found by adding the

contributions from all trajectories which pass through the beam in a given time.

Figure 4-2, which is taken from [77], shows cross sections of the laser beam and

54
 y y

P Vr

z
Zx

Figure 4-2: Cross sections of the laser beam and the trajectory of an atom in the
2 + 2 = 2 +v
x - y plane. The atom's motion is given by x2+y2 p2+vt2 where vt is the atom's
velocity in the x-y plane, and the atom passes within a distance p of the focus at
t = 0. At t = -oc the atom is well outside the laser and as it enters the beam, the
strength of the interaction with the light field grows and then diminishes as the atom
moves away.

the atom trajectory in the x - y plane. We make the approximation I,(R) = 12 (R)
I(R), and the laser beam profile is given in Eq. 4.5.
In the experiment, the axial length scales of the laser profile and the atom cloud
are typically a few centimeters. This is about 100 times greater than the radial length
scale set by the laser radius, wo. So very few atoms move any significant distance
along z while they move radially through the laser beam. Thus, we can neglect axial
motion. We also neglect the presence of the trap, and assume atoms pass once through
the laser along a straight path. This is a good approximation when the radius of the
thermal cloud is much larger than the laser radius, and the trap oscillation frequencies
are much smaller than the spectral resolution of the experiment. If these conditions
do not hold, then the spectrum will be motionally narrowed (See Sec. 4.4.5).
When an atom passes through the laser beam, the probability that it is excited
to the 2S state is given by the square of the coefficient

C2 (oo) = - f
h e-f dt"V2s,is(t")e-iws2s(t"). (4.21)

For V2s,is we take the Doppler-free term in Eq. 4.8. If we substitute the details of

55
the trajectory into this expression, we find

Cfj(oo, p,1 t,w) =d hQ(0) 2(p 2

+ vit2 )- e-i(ws2s-2)(t")
-W2
00 ( 2Z)

(4.22)
This expression shows that the excitation amplitude is the Fourier transform of the
time profile of the perturbation. The excitation probability has a Gaussian lineshape,

r Q2(0) w2 42 (2w - WIS-2S )2 2

(z)
P(oo, p, vt, w) = C (2 20 -- e (z) exp 2
8 (1 + Z2 V2 4v 2
(4.23)
The spectral width is given by the inverse of the time the atom is in the laser beam.
On resonance, the probability of being excited goes as the inverse of v2, or the inverse
of the kinetic energy.
To calculate the number of excitations per second in the entire sample, one must
find the flux of atoms with a given vt crossing the beam, which is f (vt)vtn(z), where
f(vt) is the relative probability for an atom to have velocity vt. We can take f as the
two dimensional Maxwell-Boltzmann distribution function, 2vt/u 2 exp(-v /u 2 ), and
the density, n(z) varies along z. Here, u = V2kBT/m is the most probable atomic
speed. The integral over z, p, and vt of the flux times the probability for the atom to
be excited during the trajectory yields the excitation spectrum. One finds [77]

I00'0 0fOC 2

S(v) = dp dz dvt 2 2 e-u2 vtin(z) P(oo, p,vt,w)

00 F _ _2_ _

72 0(0)wW0 d
~O1JWOJ dz n(z)
0 exp
exp) - V - "IS2 2S 147rw (z) (.4
. (4.24)
16 U - w(z) U

If the sample is confined to a region of length L < zo, in which the beam radius is
constant, the spectrum is given by

number of atoms

(0)7rWo fWd Iv 1S-2s 47rwo

S(v) = dz n(z) exp [2 . (4.25)
excitation rate per atom

This double exponential is the time-of-flight lineshape which is also observed in

56
Doppler-free two-photon spectroscopy in an atomic beam [76, 77]. The 1/e half
width is
AV = 2kBT 1 10V17-/wo Hz m K-1/2. (4.26)
m 4-Fw 0
Assuming that the laser beam waist is known, the 1/e half width is a measure of the
sample temperature.
If the sample extends beyond a divergence length from the focus, the lineshape
given by Eq. 4.24 will still approximate a double exponential, - exp(-fAvj47rwej /u),
but the effective beam waist of the spectrum, we!f, will be greater than wo due to
the divergence of the laser beam.

Experimental Observation

It was experimentally shown [10] that Eq. 4.26 is valid for the trapped sample, but
for the studies reported in this thesis, the laser waist profile is not well known. The
optical alignment was drastically changed after the vacuum system and cryostat were
assembled around the apparatus, making it impossible to observe the laser geometry in
the region of the trap. In addition, in order to produce a standing wave in the trapping
region, the beam fell near the edge of a lens and a mirror in the cryostat, so the beam
was most likely highly astigmatic. The spectra still exhibit a time-of-flight lineshape,
but the effective beam waist must be calibrated with independent determination of the
sample temperature. The temperature can be found from bolometric measurement of
the energy distribution (Sec. 2.5), the width of the Doppler-sensitive spectrum (Sec.
4.2), or a thermodynamic model of heating and cooling in the sample[24].
The optical layout was designed to produce a 47 pm beam waist, but the observed
value of the effective waist ranged between 20 and 50 pm. It would remain stable as
long as the laser alignment was not changed.
Figure 4-3 shows typical time-of-flight spectra recorded in the course of one
evening. For samples with temperatures between 100 pK and 1 mK, the temper-
ature is found with the bolometric technique. This implies an effective beam waist of
20 pm. To find the effective beam waist for warmer samples, the bolometric technique

57
 v 1.2 mK, 0.13 mK, 6.3 kHz
2500 o 5.2 mK, 0.52 mK, 11.1 kHz
o 27 mK, 2.3 mK, 19.7 kHz
^ 83 mK, 7 mK, 32.0 kHz
'n

2000

0
I)1500
(/)
00
751000
C/-)

C, 500

-60 -40 -20 0 20 40 60

Laser Detuning [kHz at 243 nm]
Figure 4-3: Typical Doppler-free spectra showing the dependence of linewidth on
sample temperature. The trap depth, sample temperature, and 1/e halfwidth are
given. The laser power in the trap region was about 7 mW and samples densities
were 1012-1013 cm 3 . The baseline of each spectrum is offset for clarity.

is not reliable. One can calculate the temperature with the thermodynamic model,

however, and the inferred waists are substantially larger than 20 pm. This is reason-

able since the warmer samples extend well beyond a divergence length from the beam

focus. If the beam shape were better known, this hypothesis could be quantitatively

verified by simulating the lineshape through numerical integration of Eq. 4.24.

Once the waist is calibrated, the time-of-flight spectrum is a valuable relative

measure of sample temperature because the /T linewidth dependence and high signal
to noise ratio extend to temperatures which are both lower and higher than the useful

regimes of other techniques.

The usefulness of the time-of-flight spectrum for determining the sample temper-
ature is demonstrated in Fig. 4-4. The effective beam waist for this data set was

determined to be 20 pm, and it is possible to measure temperatures down to 20 PK.

This data set was not recorded for the purpose of measuring time-of-flight linewidths

58
 8 I V I | 0 9 I III
r"

(a)
6

Cn
rv) - I - I
II II
2~ 140 I I I I |
120 (b)
100

80

60
I
40 0 Time-of- ight Width
0 Bolometer, Evaporation iodel
20 -U
Doppler h

0 ' I a I I I I

100 300 500 700 900 1100

Trap Depth [AK]
Figure 4-4: Plot (a) shows the 1/e halflinewidths of Doppler-free spectra from cold,
low density samples (< 1013 cm- 3 ). Each data point represents a different load of the
magnetic trap and evaporation sequence. Plot (b) shows the temperatures implied
by the linewidths and Eq. 4.26, assuming a 20 pm beam waist. This value of the
beam waist was set by independent determination of the sample temperature using
the bolometric dump method, a thermodynamic model of the sample [24], and the
Doppler-sensitive spectrum linewidth. It is possible to infer the sample temperature
down to around 20 pK.

59
and much improved data could be attained with the current apparatus. Also, more
detailed modeling of the lineshape and study of the laser spectrum is required to
better understand the very low temperature lineshapes.
For cold enough samples, as the atoms settle into the bottom of the trap, the
sample radius will become smaller than the laser radius. In this case the motional
broadening is suppressed. This is the two-photon analogue of Dicke narrowing [81]
and the M6ssbauer effect [82] in one-photon spectroscopy. In this limit, the atomic
motion is negligible and the excitation spectrum linewidth is determined by the laser
spectrum (Sec. 4.3.1), independent of the sample temperature. The data shows no
evidence that this limit has been reached since the linewidth continues to drop as the
trap depth decreases.
On the Doppler-free resonance, peak signal count rates can approach 30 counts
per 400 ps laser pulse. Assuming a 10- detection efficiency, this implies about 3
million 2S atoms per pulse. These peak rates are seen for cold samples with high
density, but low atom number (~ 1010). On resonance, for such a sample the lifetime
due to laser excitation is about 1 s, which can be short on the time scale of the
experiment. For most data discussed in this thesis, however, the excitation rate is
lower and the number of atoms in the cloud is higher, and the loss due to excitation
is not a significant factor.

4.4 Doppler-Free Excitation: Numerical Simula-

tion of Complicated Lineshapes

When the time-of-flight linewidth is small, other processes can be important to the
Doppler-free lineshape. The actual atomic trajectories in the trap can affect the spec-
trum, especially when the atoms maintain coherence with the light field over many
passes through the laser. As will be described in Chap. 5 and 6, atomic collisions
shift the energy levels so that the transition frequency depends on density. In these
experiments the axial length scale for variation of the sample density profile and

60
the laser divergence length are roughly equal, so the spectrum normally reflects the
inhomogeneous axial density and beam profile. Also, the AC Stark shift, photoion-
ization, residual Zeeman shift, and saturation of the transition sometimes need to be
considered. This section describes a semiclassical numerical calculation of the pho-
toexcitation spectrum, based on the evolution of the single-particle density matrix,
which can take all these effects into account.

4.4.1 Effective Two-Level Hamiltonian and Evolution of the

Single Atom Density Matrix

We define an effective two-level Hamiltonian for spectroscopy of the 1S-2S system,

which includes the unperturbed energies, any local level shifts (AE(r)), and the
interaction which gives rise to Doppler-free two-photon excitation (Sec. 4.1.2),

h WiS- 2 s+ AE(r) Qo(r)e2 iwt

H = , (4.27)
-2S - Z\E(r)J

The equation of motion for the density matrix, p, is

1 d
-[H p] + -P re . (4.28)
th dt

We define

P2s,2s P2s,s 1 (4.29)

P1s,2s Pis,is j

and the relaxation term reflects the laser linewidth, radiative decay, and photoioniza-
tion of the 2S state, and is given by

d -(72s + 7photo(r))P2S,2S - [laser + 72S + 7photo(r))P2S,1S

jP 0rel / (4.30)

We assign the system a peak density, no, a well defined temperature, T, and a

61
trapping potential, U(r). These imply a position and velocity distribution for the
sample, f (r, v). (f d3 r d3 v
f (r, v) = N is the number of atoms in the trap.)
For a given laser frequency, v, we calculate the contribution to the spectrum from
an atom initially in the IS state with initial position r and initial velocity v. For
a laser pulse of length tiaser, the contribution to the spectrum from such an atom
is proportional to the probability of the atom being in the 2S state at the end of
the pulse, p2s,2s(v, r, V, tiaser). This is found by numerically integrating Eq. 4.28
and Hamilton's equations for classical motion of an atom in the trap. The photo-
excitation spectrum for the entire sample is found by numerically integrating over all
initial conditions

S(v) = d3r d 3p f (r,v)p2s,2 s(v, r,v, taser). (4.31)

Equation 4.31 is a six-dimensional integral over the solution to coupled first order

differential equations. Fortunately, proper use of symmetries and approximations can

reduce the calculation time significantly. Further details of the numerical analysis are

presented in App. B.

4.4.2 Cold Collision Frequency Shift

A shift in the transition level spacing which is linear in sample density has been

observed (Chap. 5). This effect probes the quantum mechanical properties of the

gas (See Sec. 6.3), so it is not possible to correctly include it in the semiclassical

simulation of the spectrum. An approximate treatment of the shift's effect on the

spectrum is obtained by including a local level shift in the Hamiltonian,

AEcrt(r) = hXis(r) (4.32)

where nis(r) is the local density of 1 atoms and x = -3.8 ± 0.8 x 10-10 Hz cm 3 .
An atom's energy levels are shifted due to interactions with neighboring atoms.

The interactions can be represented by a mean field energy or through collision

62
based formalism and the result for a noncondensed, homogeneous sample is AEc 0 1 =
2
8-rh nis (as-2s - ais-is), where als-2s and ais-is are the s-wave elastic triplet scat-

tering lengths for 1S-2S and IS-IS collisions. The theory is discussed in Chap. 6.
A broadening is predicted to accompany the shift (App. C). It arises from dephas-
ing collisions which for a homogeneous system occur at a rate Ycol = 87ra S 2 s/i
where V is the thermal average velocity, 8kBT/7rm. The broadening at the sum fre-
quency is 7coai/27r ~ 7 x 10-9 nT Hz/v Kcm- 3 . The importance of the broadening
decreases with decreasing temperature. At 100 pK, F 0 1/27r = 6.9 x 10~" n Hz/cm- 3,
which is small compared to other linewidths in the spectrum. There is presently no
experimental observation of the broadening.
The experimental observations and simulations of spectra which are dominated
by the effects of collisions are described in Chap. 5.

4.4.3 Additional Sources of Spectral Broadening

In this section we briefly describe additional sources of spectral broadening which can
be included in the effective two-level Hamiltonian.

Laser Linewidth

The laser linewidth is 1 kHz or below at 243 nm and is usually unimportant. However,
it does set the lower bound to the width of any feature in the spectrum. For very cold
samples, when the time-of-flight linewidth becomes similar to or less than the laser
linewidth, it is not possible to use the width to measure sample temperature[10] and
the spectrum reflects the laser frequency spectrum (See Sec. 4.3.2).

Photoionization

A laser photon can excite an atom from the 2S state to the continuum, leading
to photoionization. In the laser beam, this one-photon process can proceed rapidly
because we produce a high laser intensity for the two-photon excitation. From the
calculated photoionization cross section cphoto = 7.9 x 10-18 cm 2 [83], one finds the

63
photoionization rate

rYhoto(r) = 9.7[Ji(r) + 12 (r)] s-(W/cm2)1. (4.33)

Typical laser intensities in the beam focus can exceed 100 W/cm 2 , for which 7photo(r) r

2000 s-1. This limits the lifetime of a 2S atom to 500 ps if the atom is in the laser

focus during the entire laser pulse. Since most atoms spend a significant amount of

time out of the laser beam, photoionization is suppressed and it only serves to reduce

the spectral contribution from the lowest energy atoms, rounding out the peak on

center of the time-of-flight spectrum.

AC Stark Shift

Since the electric dipole Hamiltonian associated with the photon-atom interaction

connects the S and P states, there is a significant transition frequency shift associated

with the AC Stark effect. The shift can be found from an extension of the analysis

in Sec. 4.1.2. One must calculate the level shifts that arise from transitions IS -+

nP -± 15 and 2S -+ nP -* 2S [79], which were neglected. The result is

AEAC Stark(r)/h = 3.34VII(r)12 (r) Hz (W/cm2)-1. (4.34)

In these experiments, this can shift the resonance hundreds of Hz at the laser focus.

2S Radiative Decay and the DC Stark Effect

In the absence of any perturbations, the natural 2S -+ 1S radiative decay channel[66]

is via two photons at a rate of y2s = 8.23 s-1 [84]. The resulting natural linewidth of
the transition is 1.31 Hz at 122 nm.

A DC electric field, 8, mixes the 2S level with P levels just as an AC field does.

When the excited state wave function has some P character, it can decay to the IS

state via the emission of one 122 nm Lyman-alpha photon. This reduces the lifetime

64
to[85]
T2S(E) = 72p (475/S) 2 (V/cm) 2 , (4.35)

where T2p = 1.6 ns is the 2P lifetime. In a metal-coated trapping cell, it was shown
[10] that the 2S lifetime could be comparable t o = 121.5 ms, but the cell which
-20s

was used for the experiments described in this thesis was made of plastic and the
2S lifetime was typically a few milliseconds, indicating the presence of stray electric
fields of about 0.5 V/cm.
There is also a small shift of the 1S-2S transition frequency associated with the
DC Stark mixing [59],

AEDC Stark/h 3600 .2 Hz (V/cm)~ 2 . (4.36)

Residual Zeeman Shift

The energies of the iS and 2S states shift strongly with magnetic field. This provides
the trapping potential for the atoms, and for a sample at temperature T, d state
atoms are confined by fields of B ~ kBT/B = 1.47 T T/K. Normally this makes
high resolution spectroscopy impossible for magnetically trapped neutral particles
because of the large Zeeman shift and broadening of the spectrum. The IS and
2S F = 1, mF= 1 states, however, shift identically with field, except for a small
relativistic correction [85]. The residual Zeeman shift for the IS - 2S transition in a
magnetic trap is

A Ezeeman(r)/h = B(r)a2 1 B/4h = 1.8 x 105 B(r) Hz/T. (4.37)

This effect is negligible at present. For example, for a 100 /uK sample the trapping
fields are about 1.5 x 10-' T, and the resulting frequency shift is about 30 Hz at 122
nm.

65
4.4.4 Numerical Calculation of the Time of Flight Lineshape

A check of the numerical simulation of the spectrum is to derive the experimentally

important time-of-flight lineshape. For this we restrict ourselves to the regime in
which there are no significant broadenings or shifts other than the finite transit-time
broadening. When the laser waist is much smaller than the sample radius, and the
length of a laser pulse, tiaser, is such that an atom makes one complete pass through
the Gaussian laser beam during the excitation, the probability of exciting the atom
to the 2S state should have the same form as the analytic expression, Eq. 4.23,

m2 -(27)2 _' VIs-2s )2W2 (Z)

P2S,2S(V, Z, V, tiaser) Q 2()exp V22 . (4.38)

4 - o v=91 cm/s (50 AK)

v=128 cm/s 100 AK)
v=182 cm/s (200 AK)
- --- time-of-flight (100 AK) 1

0_

0
I I I I

-30 -20 -10 0 10 20 30

Laser Detuning [KHz at 243 nm]
Figure 4-5: Probability for excitation to the 2S state for atoms which start outside the
laser beam and make one pass through the beam with zero impact parameter. The
velocities quoted in the figure refer to the velocity of the atom as it passes the origin,
and the temperature is given for which this velocity is the most probable thermal
velocity, 2kBT/m. The data points are the results of the numerical simulation for
a laser beam waist of 20 um, 1 mW laser power, and atomic motion in the trapping
potential. The solid lines are Gaussian fits. The dashed line is the time-of-flight
lineshape for a 100 pK sample.

66
 1.2 1
- Excitation on Resonance

1.0 ----- Excitation Detuned F1/

--- Excitation Detuned 2F

0.8

0.6

0.4 / .

0 0.2

0 - -
I I I I

0 0.5 1.0 1.5 2.0 2.5 3.0

Velocity/F(2kBT/m)

Figure 4-6: Contribution to the excitation rate from all the atoms with a given atomic
velocity in the laser beam, for various detunings from resonance. The total excitation
rate for a given detuning is found by integrating the area under the curve. The
curves are normalized to unity excitation rate for zero detuning. The 1/e halfwidth
is denoted F1/e.

Here, v is the laser frequency, and z and v are the atom's axial position and velocity

when it passes through the beam.

Figure 4-5 shows the behavior of the numerically calculated P2s,2s(v, Z, V, tiaser)
for atoms which are in the trap and make one pass through the laser at z = 0. At the

high atomic energies used, the velocity is nearly constant as the atom passes through
the laser. The numerical result for the excitation spectrum is described exactly by

Eq. 4.38 including the values of the width and height. The exponential time-of-flight
lineshape for a 100 pK sample, which is the weighted sum of Gaussian curves, is
also shown. The dominant contribution on line center comes from atoms with energy

substantially below kBT.

The contribution to the excitation rate, as a function of atomic energy, or peak
velocity, is shown in Fig. 4-6 for various detunings from resonance. The flux of atoms
passing through the laser with a given velocity is approximately proportional to the
three-dimensional Maxwell-Boltzmann velocity distribution, ~ v 2 exp [- (v 2 /U 2 )] (Eq.

67
4.24, Eq. B.14). This distribution function vanishes at zero velocity, which implies
that for zero detuning, a large fraction of the signal comes from a relatively small
number of low energy atoms which are in the laser for a long time.

4.4.5 Coherence Effects

When the atoms pass in and out of the laser beam more than once during one laser
pulse, the spectrum can be much different. The collision cross-section for hydrogen
is so small that there are essentially no collisions during the excitation pulse and
atoms maintain coherence with the laser beam as they oscillate in the trap. This
produces interference fringes, or sidebands on the spectrum, as observed in 1995 [10].
Numerical simulations illustrating the effect are shown in Fig. 4-7.
One can describe the excitation as a form of Ramsey separated oscillatory fields
spectroscopy [86] in which an atom passes through an interaction region at regular
time intervals. The fringe width is the inverse of the total time between the first and
last pass. The fringes lie under the time-of-flight envelope whose width is the inverse
of the time to pass through the laser once.
Another way to explain the fringes [60] is to quantize the motional states of atoms
in the trap. This picture emphasizes the analogy with sidebands observed in the
absorption spectra of trapped ions [87] and in the spectrum of scattered light from
neutral atoms in optical lattices [88]. When an atom make an electronic transition
from IS to 2S, its trap state can change if it gets a momentum kick from the momen-
tum in the Fourier transform of the spatial profile of the laser beam. The spectrum
takes the form of a carrier with sidebands. For a harmonic trap, the resonance con-
dition is satisfied for 2hv = hvis,2 s + 2
phutrap where Virap is the frequency of the
atoms orbital motion and p is an integer. Because of the cylindrical symmetry of the
system, the initial and final harmonic oscillator quantum numbers must differ by an
even number of quanta. At the laser frequency, the sideband spacing is equal to the
trap oscillation frequency. The width of the individual peaks is given by the inverse of
the shortest coherence time in the system. This time may be the inverse of the laser
linewidth, the length of the laser pulse, or the time between collisions, for example.

68
 0 5
SEE 4 100 AK Thermal Average

E
o 0
-20 -10 0 10 20

3
Single Atom
2 91 cm/s, (50 AK)

E 0
2 1.5
Single Atom
2> 1.0 - 12 cm/s, (100 AK)

0.5 -
0

-4 0.6 -
'OSindle Atom
i 0.4 - 16E cm/s, (200 AK)

0.2 -
0
-20 -10 0 10 20
Laser Detuning [KHz at 243 nm]
Figure 4-7: Numerical calculation of the probability for excitation to the 2S state after
a 1 ms laser pulse. Atoms move at z = 0 in the trapping potential and pass through
the laser beam multiple times, giving rise to interference fringes in the excitation
probability. The fringe separations are equal to the atom oscillation frequencies. The
widths result from the finite length of the laser pulse. The envelopes of the curves
are the Gaussians (Fig. 4-5) observed for the spectrum after one pass through the
laser. The velocity quoted in the figure refers to the velocity of the atom as it passes
through the origin, and the temperature is given for which this velocity is the most
probable thermal velocity, V2kBT/m. The upper trace is the excitation spectrum for
a 100 pK sample which is the weighted sum of curves such as the ones shown in the
lower three traces. The laser beam waist is 20 pm, and the power is 1 mW.
69
 1000

X 800
U) T

600 -

U400
C I

C/)
200

0
-15 -10 -5 0 5 10
Loser Detuning [kHz at 243nm]

Figure 4-8: Spectrum of a 275 pK sample showing coherence sidebands. The asym-
metry arises from the cold collision frequency shift associated with the inhomogeneous
density. Peak sample density is around 10" cm- 3 . The cold collision frequency shift,
laser linewidth of about 1 kHz, and variation of trap oscillation frequency with atom
motional energy and axial position reduce the contrast of the fringes. The dashed line
is the result of the numerical calculation of the spectrum which takes these effects
into account.

To understand all the details of Fig. 4-7, it is helpful to discuss the atomic motion
in more detail. We can neglect the axial motion, so each atom is associated with
a z position for which the profile of the trap along r is U(r) =- Bo(z) 2 + (rB')2
(Sec. 2.2.1). U(r) is nearly harmonic for r < B0 /B' and approximately linear in r
at greater distances. Thus low energy atoms exhibit simple harmonic motion with
frequency vltrap(0) = pt/mBo B'/27r. Higher energy atoms exhibit periodic motion,
but with an increasing period. The frequency for passing through the laser approaches
v"trap(E) ~ pB'/(4 /2mE) for high energy. To provide axial confinement, Bo(z) varies
with z, so Vtrap(0) also varies in the trap.
Since Vtrap varies, so does the sideband spacing. This is evident in the calculated
spectra for various atomic energies with z=0 (Fig. 4-7). The spacing decreases for
higher energy atoms and this dispersion washes out the sidebands in the spectrum for
the thermal sample. Additional dispersion, not shown in the figure, arises because a

70
real spectrum has contributions from various values of z. The cold collision frequency
shift can also obscure the sideband structure because atoms with different trajectories
in the trap see different densities and experience different shifts (Chap. 5).

Experimental Observation

In order to resolve the sidebands, their spacing must be greater that the spectral
resolution, which is ultimately limited by the laser linewidth (~ 1 kHz), and the
dispersion of the spacing must not be too great. Typically, for traps for samples with
temperature below 200 pK, the frequency vtrap(O) is between 1 and 4 kHz.
The best resolved sideband spectrum to date is shown in Fig. 4-8. The variation
of vtrap along z and the cold collision frequency shift make the experimental spectrum
less clean than Fig. 4-7. With a narrower laser linewidth, one could use a more
harmonic trap and tolerate the lower vtrap(0). With greater fluorescence detection

efficiency, one could work with lower density. These improvements should increase
the contrast of the fringes in the spectrum.
Recently, the group of T. Hinsch observed similar 1S-2S spectra using time-
domain Ramsey spectroscopy of the 1S-2S transition in an atomic beam [64].
As a final note on the subject, when the atoms are confined by the trap to within
the laser beam, there are no sidebands, just as the time-of-flight linewidth vanishes
as discussed in Sec. 4.3.2.

71
Chapter 5

Cold Collision Frequency Shift:

Observations

When the peak sample density, no, is greater than about 1013 cm- 3 , the cold collision
shift of the transition frequency is clearly visible in the spectrum. Figure 5-1 shows
spectra for a sample with an initial no = 5.0 x 1013. This chapter contains data on
the cold collision frequency shift observed in the Doppler-free 1S-2S transition, along
with simulations of the spectra and a detailed description of how we calibrate the
shift for a given density and derive the 1S-2S triplet scattering length. The result
for the scattering length is compared with a recent theoretical calculation, and the
usefulness of 1S-2S spectroscopy as a probe of trapped hydrogen is discussed.

5.1 Data

5.1.1 Experimental Procedure

Hydrogen atoms are trapped and evaporatively cooled as described in Sec. 2. For
identically prepared samples, the reproducibility of no immediately after the forced
evaporation ends is on the order of a few percent and it is measured using the bolome-
ter as described in Sec. 2.5.2.
A representative sample and trap has ~ 1011 atoms with T = 120 pK and no

72
 I I I I I I I

300

2 250
0

~200 -

0 UU,

-25 -20 -15 -10 -5 0 5 10

Loser Detuning From Unperturbed Line Center [kHz at 243 nm]

Figure 5-1: Spectra of a 120 pK sample with an initial peak sample density of
no = 5.0 x 1013 CM-3. The furthest red-shifted line was recorded first. The sig-
nal becomes less intense and the shift decreases as the sample density drops. Each
data point represents 30 ms of laser excitation. Approximately 45 seconds elapses
between the first and last scan shown. The origin of the frequency axis is the unper-
turbed transition frequency. The smooth lines are the results -ofnumerical simulations
of the lineshape.

5 x 1013 CM-3 . The axial bias field is 0.4 x 10-4 T and the axial oscillation frequency
is 10 Hz. The radial linear field gradient is 229 x 10-4 T/cm. The radial oscillation
frequency near the trap minimum is 4000 Hz. The thermal energy equals the confining
potential at characteristic dimensions of the sample, Zthermat = ±3.8 cm and rthermal =
100 pm. The laser beam waist radius is wo = 45 pm and the divergence length is

zo = 3.0 cm.

Spectra are recorded as described in Sec. 3, beginning within a few seconds after
the forced evaporation ends. While the laser illuminates the gas, helium gas evap-
orates off the retroreflecting mirror at the bottom of the cell and knocks atoms out
of the trap. The sample decays with a ~ 10 second time constant. For typical sam-
ples, about one second is required to record a spectrum with a sufficient signal-noise
ratio to determine the line center to ~ 100 Hz. With a single load of the trap and
evaporation, up to 30 useful spectra are recorded.

73
 The first scan after the forced evaporation ends probes the highest density and
shows the greatest red-shift. For subsequent scans, as the density drops, the shift
decreases. The time scale of the measurement is short compared to the time scale
for drift of the reference cavity mode to which the laser is locked (Sec. A.2.1), so the
later, low density spectra, provide the unperturbed transition frequency. Thus one
can determine the value of the frequency shift for a given spectrum without knowing
the absolute frequency of the line center.

5.1.2 Data Analysis

An example of the data analysis is shown in Fig. 5-2 for a 120 ptK sample with an
initial density of 6.6 x 1013 cm-3. The upper left graph shows the integrated signal
for each sweep and the fit to a model of a one body loss rate. Using this fit and the
initial sample density measured with the bolometer, one can assign a density to each
sweep. The first sweep is typically discarded from the analysis because the laser is
not well aligned during that time.
The upper right graph shows the 1/e halfwidths of double exponential fits to the
lines. The high density spectra are broadened due to the inhomogeneous density
distribution. This point will be discussed in Sec. 5.1.3. A phenomenological model
is fit to the curve of linewidth versus sweep number. The low density limit of the fit
gives the time-of-flight linewidth, which, along with the energy distribution measured
with the bolometer (Sec. 2.5.1) and the thermodynamic model of heating and cooling
[24], is used to determine the sample temperature and monitor the effective laser
beam waist (Sec. 4.3.2).
The center frequency of each sweep, measured with respect to a transmission mode
of the optical Fabry-Perot reference cavity, is plotted in the lower left graph. The
frequencies have been corrected for drift of the frequency of the transmission mode
(Sec. A.2.1) due to fluctuating light power in the cavity and long term relaxation of
the cavity mirror spacing. The scatter in the data is greater than the statistical errors
for the determination of the line centers, and may arise from laser power fluctuations,
changes in the laser beam alignment, or shifts in offsets in the frequency control

74
 P1 8.365 ± 0.7381E-02
0.5032E-03
i P1
P2
2.364 ±
2-462 ±
0.2528
0.2153
.} 3500 P2 -0.8129E-01 ±
5 P3 0.7186E-01 0.1634E-01

3000 CO
4.5
z
. 2500 0
4
0
2000 00

3.5
1500

1000 3

500 2.5
00
0
0 10 00 20 30 0 10 20 30
00
sweep number sweep number
I I
0
i 18 000
i 18
P1
P2
17.79
-1.319 ±
03491 E-01
01OE0

17 00 17
CU 00 CU
16
e
16 -

& 15 0 10 20 30 & 15

14 -o o sweep number
14
-0 0
13 13
-0
12 12
0 0 2

swe0ubr 0
0 11 11

10 10
0 2 4 6

density (10e1 cm-3)

Figure 5-2: Analysis of spectra from a 120 pK sample with an initial density of
6.6 x 1013 cm- 3 . Upper left: The integrated signal of each sweep. Upper right: The
1/e halfwidth of a fit of a double exponential to each spectrum. Lower left: The
center frequency of each sweep, measured with respect to a transmission mode of the
optical Fabry-Perot reference cavity. Lower right: Line center as a function of the
sample density. The slope of this linear fit is k = -1.32 kHz/(10 13 cm- 3 ).

electronics. The scatter is not a limitation at present, but one could probably control
these parameters more carefully to improve the quality of the measurement.
Finally, the line centers are plotted against the sample density obtained from the
fit of the integrated signal. We make a linear fit to this data. The slope of this line is
a measurement of k, the proportionality between the observed shift of the line center
at 243 nm and the peak sample density.

75
5.1.3 Inhomogeneous Broadening and Shift

If the region sampled by the laser were small on the length scale of density variations
in the trap, the line shape would not change as the density dropped; it would only
shift. In actuality, however, the laser samples a distribution of densities in the trap
and the lineshape and k depend on the trap geometry and sample temperature.
The effect of the inhomogeneous density distribution is evident in the plot of
the linewidths in Fig. 5-2. As discussed in Sec. C.4, the homogeneous broadening
associated with the cold collision frequency shift is small; the observed broadening is
inhomogeneous. Different atoms in the laser beam see different average densities and
experience different shifts of their resonance frequency.
The physically interesting cold collision shift parameter is x, which is defined by
the local mean field relation between the level shift and the local density, AEis- 2s(r) =
hXn(r). The relation between x and k is X = 2k/a, where the factor of two refers
the shift to the 1S-2S energy level spacing at 122 nm, and a, which varies with the
experimental geometry, corrects for the inhomogeneous density distribution in the
trapped gas. For every trap and sample configuration, a is determined by numeri-
cally calculating the spectrum (Sec. 4.4), including the cold collision frequency shift
and the density distribution and laser spatial profile. Then, for every k, one can
extract a measurement of x.
The smooth curves in Fig. 5-1 are the results of simulations and demonstrate that
in this regime the calculation agrees with experiment reasonably well. It is found that
for a given trap and sample temperature, the shift of the line center is a constant
fraction (a) of the shift associated with the peak sample density, independent of
the peak density. For different magnetic trap profiles and temperatures, the fraction
ranged from 0.7 to 0.8. For a discussion of some the limitations of the numerical
simulation of the spectrum, see Sec. 6.3.

76
5.1.4 Systematic Uncertainties

The statistical uncertainties are small for an individual measurement of X. There are
significant possible systematic errors however.

Density Calibration

As discussed in Sec. 2.5.2, we determine no by measuring the dipolar decay rate

through bolometric detection of the atoms after they are released from the trap. The
dominant uncertainty in density measurements arises from imperfect knowledge of our
trapping fields on the 10-4 T level due to trapped fluxes in our large superconducting
coils and fringe fields from the 4 T field in the discharge region. This produces
systematic errors of about of 10-20%.
Any error in the calculated value of g would also be reflected in the value of X.

no(10 13 cm- 3 ) T(pK) k(kHz/(10 13 cm- 3 )) a X = 2k/a(kHz/(10 13 cm- 3 )

3.1 ± 0.3 490 -1.59 ± 0.16 0.79 -4.02 ± 0.80
3.1 ± 0.3 490 -1.45 ± 0.15 0.79 -3.67 ± 0.73
3.1 ± 0.3 490 -1.66 ± 0.17 0.79 -4.20 ± 0.84
3.1 ± 0.3 490 -1.75 ± 0.18 0.79 -4.43 ± 0.88
5.5 ± 0.6 250 -1.22 ± 0.12 0.7 -3.49 ± 0.70
4.5 ± 0.9 250 -1.20 ± 0.24 0.7 -3.43 ± 1.03
4.5 ± 0.9 250 -1.11 ± 0.22 0.7 -3.17 ± 0.95
4.5 ± 0.9 260 -1.84 ± 0.37 0.79 -4.66 ± 1.40
4.5 ± 0.9 260 -1.85 ± 0.37 0.79 -4.68 ± 1.40
4.5 ± 0.9 260 -1.71 ± 0.34 0.79 -4.33 ± 1.30
3.2 ± 0.6 110 -1.69 ± 0.34 0.79 -4.28 ± 1.28
3.2 ± 0.6 110 -1.66 ± 0.33 0.79 -4.20 ± 1.26
6.6 ± 0.7 120 -1.25 ± 0.13 0.8 -3.13 ± 0.62
6.6 ± 0.7 120 -1.46 ± 0.15 0.8 -3.65 ± 0.73
6.6 ± 0.7 120 -1.32 ± 0.13 0.8 -3.30 ± 0.66
6.6 ± 0.7 120 -1.51 ± 0.15 0.8 -3.78 ± 0.76
7.1 ± 1.4 120 -1.36 ± 0.27 0.8 -3.40 ± 1.02

Table 5.1: Data for determining the cold collision frequency shift. The symbols are
defined in the text. Errors in k are from the errors in no. Uncertainties for X are the
sum of a 10% uncertainty arising from a and the uncertainty coming from k.

77
Geometry Correction Factor a

The calculation of a for a given trap is most sensitive to the assumed position of
the laser focus with respect to the atom cloud. Radially, the atoms are overlapped
with the laser experimentally by monitoring the 1S-2S signal while varying magnet
currents to move the cloud. Axially, it is difficult to move the cloud and there is
an uncertainty in relative position on the order of two centimeters. By performing
calculations for various reasonable laser focus positions, it was determined that for a
given sample there is a 10% uncertainty in a.

5.1.5 Measured Value of x

To examine the sensitivity to systematic effects, X was measured in different trap con-
figurations, with sample temperatures between 110 and 500 pK and initial maximum
densities in the range (2 - 7) x 1013 cm- 3 (Tab. 5.1). For a given measurement, the
uncertainties are systematic, rather than statistical, so, to be conservative, we add

0
--
E
-2-

CD-3 I
-4
LXJ

-6 -
100 200 300 400 500
Sample Temperature [AK]
Figure 5-3: The frequency shift parameter X, determined as described in the text
for various trap configurations. No significant dependence on sample temperature is
observed. The error bars reflect systematic uncertainties in the magnetic trapping
fields and laser geometry. The dashed line is the weighted mean of all measurements,
X = -3.8 ±0.8 x 10-" Hz cm 3 . and the quoted uncertainty is indicated by the double
arrows.

78
them linearly, rather than in quadrature.
In Fig. 5-3, the various measurements are plotted versus sample temperature. We
observe no significant temperature dependence of the shift, consistent with theory.
Any significant nonlinear density dependence of the shift would manifest itself in data
such as the lower right graph in Fig. 5-2, but none is evident.
From a weighted average of the various measurements, we find X = -3.8
0.8 kHz/(10 13 cm- 3 ). Under the worst case scenario, the systematic errors could

all be of the same sign, so we give the uncertainty in X as the average of the sys-
tematic uncertainties of the individual measurements. In an apparatus optimized
for spectroscopy, rather than achieving BEC, the magnetic field and laser geometries
could be better known, and the uncertainty in X could be greatly reduced.

5.2 The 1S-2S S-Wave Triplet Scattering Length

5.2.1 Experimental Value of the 1S-2S Scattering Length

For a homogeneous sample and excitation, assuming the observed frequency shift
arises entirely from elastic collisions, X is given by (See Chap. 6),

X = (als-2s - ais-is) m (5.1)

From this, we derive als-2s = -1.4 ± 0.3 nm. We have used the theoretical value of
ais-is = 0.0648 nm[89], which constitutes only a small contribution to X.
The sample is not homogeneous, however, and possible implications of this fact
for the expression for X are discussed in Chap. 6. Since the theory for the frequency
shift in an inhomogeneous system is not completely understood, the measurement of
ais-2s should be viewed as preliminary. Further experiment and theory are needed.
It is safe to neglect contributions to the shift from inelastic processes. The most
important of these effects would be collisional hyperfine transitions and quenching of
the 2S state. Hyperfine transitions in the spin-polarized sample arise only through
weak magnetic dipole interactions[42], and in our experiment we observe no evidence

79
for collisional quenching on a millisecond time scale. Thus it is reasonable to neglect
inelastic processes when describing the observed density-dependent frequency shift.

5.2.2 Comparison with Theory

Jamieson et al.[90] calculated als-2s= -2.3 nm. No uncertainty was attached to

this value, but a rough estimate[91] is about ± 30%. With this uncertainty, the
calculation is in fair agreement with experiment.
Excited 2S atoms in the spin-polarized sample interact with IS atoms on the e3E+
potential. The theoretical result for the scattering length was based on a potential
derived from several ab initio calculations [92, 93] for small interatomic separations
and an attractive van der Waals potential for large separations. The authors comment
that the scattering length is particularly difficult to calculate because the potential
is only slightly too shallow to support an additional bound level. In addition, they
cautioned that the potentials do not mesh smoothly and somewhat arbitrary inter-
polations are required. Different interpolation methods produced 1S-2S scattering
lengths which varied by 15%. Extension of the ab intio potentials could reduce the
uncertainty in the calculation. A possible way to determine the potential to high
precision is to perform photoassociative spectroscopy of bound states of the e3 E+
potential.

5.3 Using the Cold Collision Frequency Shift as a

Probe of the Trapped Gas

5.3.1 Noncondensed Gas

With the calibration of the cold collision frequency shift, one can use the Doppler-
free 1S-2S spectrum of the noncondensed gas to determine sample density. This
proves particularly useful for colder and more compressed traps for which bolometric
determination of the density becomes unreliable due to decreasing signal strength and
increasing uncertainty in the shape of the magnetic trap. When the sample radius

80
 5 I5 I I 1

u)*34

X V

C')

U)/U

0
-60 -40 -20 0 20
Laser Detuning [kHz at 243 nmn]

3 2 1 0
Density Corresponding to Frequency Shift [x1 4 cm-3]
Figure 5-4: Spectra of a 55 pK sample (trap depth = 350 pK) with initial peak sample
density of between (1 ~ 2) x 1014 cm-3. The most intense spectrum was recorded
first and the signal strength and density are lower for subsequent traces. The time-
of-flight 1/e halflinewidth, measured from the low density spectra, is 2.2 kHz and
is much smaller than the widths of the high density spectra, which arise from the
inhomogeneous density sampled by the laser. The unperturbed center frequency of
the low density sweeps is taken as zero detuning. The density which corresponds to
a given frequency shift (n = 2Av/x), is indicated below the graph. Each data point
corresponds to 12 ms of laser excitation. The dashed line is a crude simulation of the
spectrum (Eq. 5.2) for a sample with no = 1.5 x 104 cm-3

is greater than the laser beam radius, which holds for all but the lowest accessible
temperatures and strongest magnetic trap compressions, the 1S-2S signal intensity
varies as the sample density. The bolometer signal, however, varies as the number of
trapped atoms. As the sample is evaporatively cooled, the atom number decreases
so the bolometer signal intensity drops, but the peak density generally increases as
atoms settle into the trap, so the 1S-2S signal strength increases.
Figure 5-4 shows spectra of a 55 pK sample (trap depth = 350 pK). The lineshape
is dominated by the cold collision frequency shift. Numerical simulations of the

81
spectra are not particularly illuminating because uncertainties in the trap and laser
geometry are significant for such a cold sample and the semiclassical calculation of
the lineshape is not expected to be accurate when the cold collision frequency shift
dominates (Sec. 6.3).
For a high density sample, a simple approximate calculation of the lineshape is
instructive. If one interprets the spectrum as a histogram of densities in the laser
beam, the spectrum is approximated by

S(v) J d3r F v - X n(r) y) n(r)12(r), (5.2)

where 1(r) is the laser intensity profile and F(v; 7) is a lineshape function with width
7, centered on v = 0. The convolution enforces the resonance condition and allows
inclusion of a laser linewidth. By allowing y to depend on local density, one can
incorporate the broadening due to 1S-2S collisions. Such a calculation neglects atomic
motion.
The result of such a calculation is shown in Fig. 5-4. It does not reproduce the
detailed shape of the data, but as a simple approximation, it shows that the histogram
interpretation is not too far off. Because the experimental lineshape is not yet well
understood (Sec. 6.3), however, a discussion of the discrepancy between data and the
simple calculation is not possible.
The initial peak sample density can be estimated from the peak shift observed
in the first recorded spectrum. One must allow for transit-time broadening, laser
linewidth, and broadening due to the 1S-2S collisions, so the initial peak sample
density determined from the spectrum is between (1 ~ 2) x 1014 cm- 3 . In an
apparatus with less uncertainty in the magnetic field and laser-atom overlap, study
of such low temperature, high density spectra could provide a great deal of information
on the cold collision frequency shift in an inhomogeneous system.

82
5.3.2 Bose-Einstein Condensation

Spectroscopy of the 1S-2S transition has proven to be an invaluable tool for studying
Bose-Einstein condensation in hydrogen. Section 6.2 derives the condensate 1S-2S
spectrum. The large density in the condensate shifts the resonance far to the red
(Fig. 6-6), and the shape of the spectrum can provide valuable information on the
distribution of densities and the condensate wave function. Experimental results are
discussed in more detail in Chap. 7.

83
Chapter 6

Cold Collision Frequency Shift:

Mean Field Theory

The pressure shift and broadening of an atomic spectrum arises because an atom's
energy levels are perturbed due to interactions, or collisions, with neighboring atoms
[14]. The thermal average phase shift per collision is q = 27a/AT, where a is the
s-wave scattering length for the collision and AT = h 2 /27rmkBT is the thermal de
Broglie wavelength. When q < 1 only s-waves are involved in the collisions and the
effect on the spectrum is often called the cold collision frequency shift.
Theory for the shift in the low temperature regime, based on the Boltzmann
transport equation (App. C), has been developed to explain observations in cryogenic
hydrogen masers [15] and laser cooled atomic fountains[18, 16]. These calculations
assume a homogeneous sample and excitation and find the shift resulting from elastic
collisions for a coherent, weak excitation in a two-level system is

87rh
27rAvco = -(a,-2 - a 1 _1 )ni, (6.1)
m

where state 1(2) is the ground(excited) state, ni is the density of the sample, and
a,_ is the s-wave scattering length for a - 3 collisions. There is also a broadening
of the transition, F/27 = n14a_2(retive) (assuming a 1 2 | |_1), which arises
from dephasing collisions. For an atom in state 2, F is the normal collision rate with

84
 800

600

02400-
C4

200
i i I I *'* .

-25 -20 -15 -10 -5 0 5 10

Loser Detuning From Unperturbed Line Center [kHz at 243 nm]

Figure 6-1: Doppler-free spectra of noncondensed trapped hydrogen, showing the

cold collision frequency shift. The sample temperature is 120 AK and the furthest
red shifted line corresponds to a peak density of ~- 6.6 x 10" cm-3. Sample density is
proportional to the integrated area under a given trace. The small extra broadening
of the high density traces is inhomogeneous, arising from the inhomogeneous density,
distribution in the trap. Each data point represents 30 ms of laser excitation and the
line center of the low density trace is taken as the frequency origin.

atoms in state 1, assuming the cross section is a- = 87ra 2 as would be the case for
collisions between identical particles. The broadening is equal to V Z#_times the shift.'

If # is large, then the line is mostly broadened. If 0 is small, however, as is the case
for the cold collision frequency shift, the atom's oscillating dipole retains its phase
relation with the electric field for many collisions and the resonance is mostly shifted.

We have observed the cold collision frequency shift in the 1S-2S spectrum of

trapped atomic hydrogen [11] for both a noncondensed sample (Fig. 6-1) and a Bose-
Einstein condensate (BEC) [13]. The atomic motion and the inhomogeneous trap
make it nontrivial to quantitatively explain the observations by generalizing the Boltz-
mann transport equation treatment. To gain some insight, and to derive expressions
for the frequency shift which are appropriate to these new experimental conditions,
we present a simple calculation based on a mean field interaction. We specifically con-
sider Doppler-free 1S-2S excitation of trapped hydrogen, but the basic results should

85
apply for any excitation scheme so long as the momentum imparted to the atoms by
the light does not change the spatial distribution of the particles appreciably during
the time of the excitation.
To illustrate some of the concepts, we will first consider a simple homogeneous
system, both for a noncondensed and condensed gas. Then we will consider the more
realistic situation of an inhomogeneous system.

6.1 Mean Field Description for a Spatially Homo-

geneous System
The gas is assumed to be homogeneous and dilute (na3 < 1), where n is the den-
sity and a is the s-wave scattering length, and one can neglect interactions between
more than two particles. Also, a < AT, which in hydrogen is satisfied below a tem-
perature of about 1 K [94], so interactions arise only through s-wave collisions. For
s-wave collisions, the interatomic potential can be replaced by a shape independent
pseudopotential[95] corresponding to a phase shift per collision of ka, where hk is
the momentum of each of the colliding particles in the center of mass frame. In this
regime, the cold collision frequency shift can be described with mean field theory[96].
The important scattering lengths are the triplet scattering lengths, ais-is and
ais-2s. We neglect 2S-2S scattering because the excitation rate is assumed low (in
the experiment typically 10-4 of the atoms are excited), so the background gas is
essentially pure IS. Collisions between IS and 2S atoms produce the dominant
effect and the observed shift is to the red.
Inelastic collisions, such as collisions in which the hyperfine level of one or both of
the colliding partners changes, will contribute additional shifts which are not easily
explained in this formalism, but these effects are small in the experiment.
We consider a gas of N particles. The translational states of the atoms are plane
wave states with periodic boundary conditions in a box of volume V. As N, V -4 00,
physical observables do not depend on the boundary conditions. We identify N/V as

86
the particle density.
The wave vectors of the atoms form an ordered set of motional quantum numbers
ki. For a spatially homogeneous sample and excitation profile, the probability of
an atom being promoted to the 2S state is independent of its motional state. This
is not a good approximation of the real experimental situation but would describe
excitation in, for instance, a hydrogen maser bulb, where the density and excitation
rate are homogeneous in the sample.

6.1.1 State Vectors

Before laser excitation, the system consists of N particles in the IS state. For the case
of a noncondensed gas, with no two atoms in the same motional state, the normalized
state vector before excitation is

n = S 1, 1 IS,k 2 ; ... 1,kN)- (6-2)

We define S as the operator which creates a normalized state which is symmetric

with respect to particle label. For state 6.2, S = 1 EQ Q, where the sum runs over
all N! permutations Q of the particle labels. For the state vector, we use the ket
notation (I...)), in which the entry in the first slot is the state of atom 1, the second
entry is the state of atom 2, etc.
For a condensate at T = 0 with N particles, all ki = 0, where 0 denotes the
motional ground state of the system. The normalized state vector before excitation
is

= ___C) 0; ... 1_,

I,1S, 0). (6.3)
N terms

To form the normalized state vector of the excited state, we note that during
Doppler-free excitation of a homogeneous system by a spatially homogeneous laser
profile, an atom's motional state is left unchanged (See Sec. 4.4.5 and [60]). Assuming
a monochromatic laser, the internal state evolves unitarily [97]. If the amplitude
of excitation for each atom is r, and the amplitude for remaining IS is t, where

87
t 2 + r 2 =1, then IIS, kZ) -> t|lS,ki) + r|2S, ki), and the excited state vector is

N N'
e) E tN-n n T N-n;n (6.4)
n=O n!(N n)

The normalized state N-n;n ) describes a gas with N - n IS atoms and n 2S atoms,
where x can either refer to the normal or condensed gas,

|TN-nn RSI27k --.. 2S, kn IS n+1; --- IS, kN)

F " = S 2S, 0; ...2S, 0; iS, 0; ...1S, 0). (6.5)

n terms N-n terms

Here, as before, S =- EQ Q for the normal gas. For BEC, S n!(N-n)!

where E' runs over the N!/n!(N - n)! permutations which create distinct arrange-
ments of the particle labels. The new operator, R, symmetrizes the state vector with
respect to the motional states, ki, which are excited to the 2S level. For example, for
N = 3 and n = 1,

'J2,1rm) = IZS2S,ki;1S,k 2 ; 1,k 3)

1
S [12S, ki; 1S, k 2 ; 1S, k 3 ) + 12S, k 2 ;1S,k 3 ; 1S, k1 )

+2S, k3 ; 1S, ki; IS, k2 )], (6.6)

and S then symmetrizes with respect to particle label. 1

In 1E), the number of atoms in the 2S state is not a good quantum number.
However, the probability distribution is strongly peaked around h = r 2 N 2S atoms,
and the excited state vector can be approximated by ) - This is
illustrated in Fig. 6-2 for a sample with N = 10'.
Both states KTnNm) and IT-n; n) are symmetric with respect to exchange of any
momentum state labels ki and kj. When all atoms are 15 this is required by the
statistics for identical bosons, but nature does not require the state vector to be

'One is not required to exchange the particle labels of IS and 2S atoms, but the physical ob-
servables of the system are unaffected by this and it is mathematically simpler.

88
 0.150 I . I I I I I I I I I I I . I I ' I '

0.125

0.100

0.075

0.050

0.025

100 101 102 103 104 105 106 10

Number of 2S Atoms

Figure 6-2: Probability distributions of the number of atoms in the 2S state for a
total of 107 atoms and various fractions of atoms excited. The left most distribution
is for each atom having a 10-6 probability of being excited (r = .001), the middle
is 10-4 (r = .01.) and is multiplied by 10, and the right most is 10-2 (r = .1) and
is multiplied by 100. The sum of the probabilities for each case is 1. Note that the
distribution is quite narrow except for very small excitation probabilities. In our
experimental situation we typically have a 10- to 10-4 excitation probability for
each atom.

symmetric under exchange of the motional states of a IS atom and a 2S atom. The
momentum exchange symmetry of 4N- ) is particular to our assumption that the
excitation probability is equal for all motional states and reflects the homogeneity of
the system.

6.1.2 Hamiltonian and Mean Field Energies

The total many-body Hamiltonian is

N 2
H i + Hin) + H', (6.7)

where P, is the momentum operator for particle i, H'nt is the Hamiltonian for the
internal state of atom i, and H' is the operator for the two-body interaction which
contains the effects of IS-IS and 1S-2S elastic collisions. Using pseudopotential

89
formalism [95, 96] H' is written

2
H' H'-4irh
=E N
M1<i<j
6(i - ij) {aislispispls + as-2s [plsp2s + p2spis] }
N

= wij. (6.8)
1<i<j

The sum is over N(N - 1)/2 distinct terms. Here, ii is the operator for the position of
atom i. The operator Pi = (Jx)(xJ)i, where x = IS or 2S, projects the internal state
of atom i onto state Jx). Note there are no terms of the form (J1S)(2SJ)i(J2S)(IS )j,
which would involve excitation exchange [98] - a distinct process from those described
in 6.8. This process could take place in the sample, but we neglect it here [99].
The energy before laser excitation (N IS atoms and 0 2S atoms) is

N h2 k 2
EN;O h2 i + E/N;O (6.9)
-=1 2m

where EIN;o is the interaction energy (See Sec. D.2),

E/N;O - (N;O|H'IN;O)
27wh 2ais-is N 2
(2)
Mm V g (0), (6.10)

where g(2 ) (x) is the density-normalized second order spatial correlation function of
the gas [100, 101] before excitation (See Sec. D.1),

(2) (X)= I d3 r (N;0- r)6(ij - r - N; (6.11)

The correlation function expresses the statistics of the system. It shows antibunching
for identical fermions (g (2 )(0) = 0), bunching for incoherent identical bosons (g(2) (0) =
2), and classical behavior for coherent identical bosons (g( 2 )(0) - 1).
The interaction energy EIN;' has the form of a mean field energy in which the in-
dividual pair-wise interactions have been replaced by an average over the distribution
of atomic positions in the sample.

90
 The total energy of the system in the excited state, jN-F;n) is

h 2k 2
N
EN-l;l = Eis-2 s + E - + EN (612)
.2m

As discussed in greater detail in Sec. 6.3, the excited state is one state in a manifold
of states with ft 2S atoms. The manifold is degenerate in the absence of H' and
one must use degenerate perturbation theory to find the interaction energy, E'N-l;.
Fortunately, the state excited by the laser (Eq. 6.5) is an eigenstate of H', and the
interaction energy of the excited state is given by (Sec. D.3)

EN-; __ N-ii;ii|H'| IN-A;A)

2h 2 N(N - i) [(N - f - 1)ais-is + 2hais-2s] g(2 )(0).

(6.13)
m V(N - 1)

For N > i, the energy to excite h atoms to the 2S state is

EN-;N - EN;O N
__ - - a13-1)g( 2 )(O) + i E 1s 2 s. (6.14)
m V

For a nondegenerate Bose gas, g( 2 )(0) = 2. From this it follows that the energy
supplied by two laser photons to excite an atom out of a normal gas is

2
Nf-;n -_ E ""
E nom N;O _ N
- 8h -(aS-2S als-is) + Es-
- 2 s. (6.15)
ni M V

For a condensate, g(2)(0) = 1, and the energy required for excitation is

EN-h;
BEC - E BEC
N; 4 h2 N
iS-2S a~s~-1s) + E 1 s- 2s. (6.16)

6.1.3 Discussion

We interpret the transition onergy shifts in Eq. 6.15 and 6.16 as arising from mean
field shifts of the energy levels, as shown in Fig. 6-3. Spectroscopy of the 1S-2S
transition measures the separation of the perturbed levels.
In Eq. 6.14 for the transition energy shift, the term arising from 1S-2S interactions

91
 AL

2S
8 h2a1S-2S n IS
AE=-
m

iS 8Ab2a-sasn
1 s
$2 m

Unperturbed Perturbed
Levels Levels
Figure 6-3: 15-2S energy level diagram for a noncondensed, homogeneous sample,
showing the density-dependent level shift which gives rise to the cold collision fre-
quency shift.

depends on the iS-1S correlation function g( 2 )(0) in the same way that the 1S-iS
interaction term does. In quantum mechanics, 2S particles are considered distin-
guishable from IS particles and one might not expect such exchange effects to appear
in the 15-25 interaction. In other words, for a given 15 density, one might expect
the same shift due to the 1S-2S interactions in a condensate or normal gas. The
dependence on g( 2)(0) for the 1S-25 interaction arises in this case from the special
form of the excited state given by Eq. 6.5. Since all motional states in the original
system are equally likely to be excited, the 2S excitation is equally shared by every
atom. Consequently, in addition to the direct contribution to the 1S-2S interaction
energy from terms of the form (2S, k,; IS, k 2 |H'12S, ki; IS, k2 ), because of the extra
symmetry in the state vector, there is also an equal contribution from exchange terms
like (2S, ki; 15, k 2 JH' 2S, k2 ; 15, k1 ) (k1 and k2 have been switched in the ket.). This
can be interpreted as implying that the motion of the excited 2S atoms is correlated
to the motion of the 15 atoms, and it is a purely quantum mechanical effect. This
correlation can be seen in the state vector for the excited state of the normal gas, Eq.

92
6.5, which is the coherent sum of state vectors of many configurations of the system.
If in one configuration, a 2S atom is found in motional state ki, then in almost every
other configuration, there is a IS atom in state ki.
If one were to measure the interaction energy of a 2S particle introduced to
the sample from far away, its motion would not be correlated with the motion of
the IS atoms. There would be no exchange contribution, and the energy would be
47rh 2 ais-2snls/m.

6.2 Mean Field Description for a Bose Condensed

Gas in a Magnetic Trap
If one makes a local density approximation, in which the excitation of a given atom
takes place in a small region in space where the shift is given by Eq. 6.16, with N/V
replaced by the local density, then the spectrum for the condensate is given by

EIS-2s+ 47h 2 (as-2s - ais-is) n'EC

6
S(v) oc d3r niEC(r) 2hBE

(6.17)

This expression turns out to be correct, but it does not provide much insight into the
excitation process. For instance, how can one make a local approximation when the
condensate wave function extends over a large region of space? We present a more
rigorous derivation of the spectrum which explicitly calculates the state vector and
energy of the system before and after excitation.
When the system is Bose condensed, the density becomes so high that the inter-
action energy significantly modifies the wave functions for an inhomogeneous system.
In this case one cannot treat the interaction as a perturbation as was done in the
previous section.

93
 The Hamiltonian for the system is now

N (p2
H= 2( +U(i) + Hnt + H', (6.18)

where U(r) is the trapping potential, which for a magnetic trap is effectively the same
for iS and 2S atoms. H' is given in Eq. 6.8. We treat the condensate in the T = 0
limit and leave finite temperature effects for a future study. We specialize to the case
of ais-is > 0 and ais-2s < 0.

6.2.1 System before Excitation

At T = 0 all the atoms are in the condensate, and the condensate state vector before
excitation is
) =iS, 0'NS; ... is oN,1S) (6.19)
N terms

where JONJS) is the single particle motional state of a IS atom in a condensate with
N atoms. It is important to note that by writing the condensate state vector as Eq.
6.19, we have set g( 2 )(0) - 1. For the system to have g( 2)(0) # 1, the state vector for
the "condensate" would have to involve more than one motional state.
One finds ON,s) by minimizing E N; = (IFNc|HJ pNc) under the constraint

of there being N total particles. The constrained minimization leads to the Gross-
Pitaevskii, or nonlinear Schr6dinger equation [102, 103, 104] for the single particle
BEC wave function, <NEC(r) = (r oN,1S). The kinetic energy is small and can be
neglected. This leads to the Thomas-Fermi wave function,

4BEC (r) = _kECr

N- 1/ 2 (rCEC() - U(r)/U) 1 /2 U(r) < nN EC(0) (.

( BE BE, (6.20)
0 otherwise

where nNEC(r) is the density distribution in the condensate, nEC(0) is the peak
density, and = 47h 2 ais-is/m. It can be shown for a cylindrically symmetric trap

94
that

nBEC(O) = 0.118 (Nm3 2rz/a s-IS)2/57 (6.21)

where Wr and wz are the frequencies in rad/s for small radial and axial oscillations
at the bottom of the trap. One can interpret ON* c(ri)NEC(r-) as the probability of

finding condensate particle i at position ri. For a harmonic trap, @NC (ri N

varies as an inverted parabola along any trap axis (See Fig. 6-4).
In the Thomas-Fermi approximation, the Lagrange multiplier enforcing particle
conservation in the constrained minimization is the chemical potential,

[-IN hWr [5Naisis mwr 1/2 z 2/5 nN (6.22)

2 h UMr B 0

and because PN OBEC/ON, the energy of the ground state before laser excitation
is
NO 5
EC _5NAN-
E~ -N-;
7-2 - (6.23)

6.2.2 System after Excitation

We consider a homogeneous excitation, which is a good approximation for the hydro-

gen experiment because the laser is uniform over the region of the condensate.
When p condensate atoms are excited to the 2S state, 0 N/EC(r) is no longer the
single particle wave function which minimizes the energy of the condensate atoms.2
For a weak excitation (p < N), we can neglect the small perturbation of the BEC
wave function due to the presence of the 2S atoms, and the new equilibrium con-
densate many body state vector is given by I NC' , as defined as in Eq. 6.19 and
6.20.
The 2S atoms interact strongly with the IS atoms in the dense condensate. To
find the effective potential in which 2S atoms move, we will see that one adds the IS-
2S interaction, AE(r) = (4rhais-2s/m)nN-C(r),to the magnetic trapping potential.

2
The interesting problem of determining the relaxation time of the wave function [105, 106] after
excitation is left to future study and we assume the excitation is slow enough that the excited state
is in equilibrium.

95
 10

'-1 -10

2 -20

a-30-

-40
-2SM
Magnetic Potential
-50 -Effe Potential

-15 -10 -5 5 10 15
8

2
-- BEC\
C/) - Magnetic Potential \/
- Effective Potential
0 I
-
I I
-.

-15 -10 -5 0 5 10 15
Radius [pm]
Figure 6-4: Radial potentials and single particle wave functions for iS atoms in the
condensate and 2S atoms trapped in the condensate interaction well. The dashed line
is the magnetic trapping potential U(r), which is identical for IS and 2S atoms. The
solid line is the effective potential in which atoms move, which includes the mean field
interaction energy. Note the different energy scales. The heavy solid lines represent
probability density distributions for atoms in various states. For the 2S states, every
third state in the BEC interaction well is shown, up to the twelfth. The light solid lines
indicate allowed Doppler-free transitions from the condensate, which must preserve
mirror symmetry. The radial potentials and condensate wave function are for a peak
condensate density of 5 x 105 cm- 3 (PN/kB : 2 pK), and a radial trap oscillation
frequency of 4 kHz, which are characteristic conditions for the MIT hydrogen BEC
experiment. The scattering lengths used in the calculations are ais-is = 0.0648 nm
and als-2s = -1.4 nm. The 2S wave functions shown are one-dimensional simple
harmonic oscillator wave functions. The bound 2S levels form a near continuum of
motional states in a realistic anisotropic three dimensional trap.

96
This forms a deep potential well in which 2S motional states are trapped. At a given
laser frequency, the condensate atoms are excited to one of the states in the well (See
Fig. 6-4).
The energy of the excited state after laser excitation, and the 2S motional wave
functions, are found by minimizing the energy functional,

gN-pp,i pXFN
H|WN C 3 (6.24)

where 3

NP;P (6.25)
kN-p, 2 S. 2, p,2S. is, 0 N-p,1S; ... 1S, 0 N-p,IS),
p terms N-p terms

and k N~p,2 s is the 2S motional state which is resonantly excited. Using Eq. 6.18, the
energy functional reduces to (See Sec. D.4)

EN-p;p,i
EBEC
4,r h 2 a 1S-2S N-p 2 -P,2S)
+p(2S, ki>P,2 s [Hint +2m
m BEC

= EN- 0 + p(EIs-2s + j), (6.26)

0 is defined in Eq. 6.23.

where E N-p
Finding the 2S motional states which minimize this functional, with the require-
ment that the 2S motional states form an orthonormal basis, is equivalent to finding
the eigenstates of the effective 2S Hamiltonian

2_
"2S
He
_
ff = 2 + U(i) + 4rh 2 ais 2 s mN- (6.27)
2m m

and the eigenvalue for state k N ,2s is Ej. The effective potential and some 2S motional
states are depicted in Fig. 6-4.
Inside the BEC potential well, using the Thomas-Fermi BEC density distribution,

3 The excited state has a distribution in the number of 2S atoms, as discussed in Sec. 6.1.1, but we
assume p is large enough that the distribution is narrow and peaked around a well defined number.

97
the eigenstates and eigenvalues are approximately those of a three dimensional har-
monic oscillator with spring constants larger than those of the magnetic trap alone
by a factor of 1 - als2s/als-s. States near the top and outside of the BEC
interaction well are more complicated.

6.2.3 Discussion

From Eq. 6.23 and 6.26, the energy supplied by two photons to drive the transition,
for p < N, is given by

2hv = EN-P'P EN 0 _ p(EIS-2S + Ei) + E G - ENC

p p
SEIs-2s + ei - PN- (6.28)

We have used (E Nc - EB
N 0
)/P ',_BEN C/ = IN for small p. Note that E& < 0
for states bound in the BEC interaction well. Since there are many 2S motional
levels which may be excited (See Fig. 6-4), there will be a distribution of excitation
energies which extends from 2hv ~ Eis-2s on the high frequency side, to 2hvp
EIs-2s+47h 2ais- 2 s nBEC(0) /m - pN on the low side, where v is the laser frequency.

The lowest frequency is for transitions to the lowest state in the interaction well. The
spectrum dies off above EIs-2s because states outside the well have negligible overlap
with the condensate and are essentially inaccessible by laser excitation.
The excitation rate when the laser is resonant with 2S motional states with energy
E, S[2hv = (EIs-2s + E - PN)], is proportional to E;,, I(kN-p,2S oNJS) 2 or

S(2hv) = hQ0(0) (kN-p,2S ON,1S) 2 6(2hv - E 1 s- 2 s - Ei + pN) (6.29)

where the delta function enforces the resonance condition. The Rabi frequency for
Doppler-free excitation in a standing wave is denoted (Eq. 4.17)

Qo (R) = 2M229
'
is
2R,,, 3 11(R),I (6.30)
37r2hC

98
where 1(r) is the laser intensity profile.
An interesting result of this calculation is that the broadening of the Doppler-free
BEC spectrum is homogeneous. Since all atoms are initially in the same quantum
state, the spectrum of every atom is identical. Recall that the Doppler-free excitation
spectrum of the normal gas is inhomogeneously broadened by the frequency shift and
the density distribution (Sec. 5.1.3 and 6.3).
Because the 2S single particle wave function N-P,2S (r) = (r~k p, 2 S) oScil-

lates rapidly, the overlap integral, (k-P 2

sON,1S), will be most sensitive to the
value of ONEC(r) =(rONJlS) at the classical turning points, which is approximately

BEC(rturn) - rnBEC(rturn) /N. Intuitively, the sum over states at energy E inte-
grates the condensate density in a shell at the equipotential surface defined by r ,urn
the turning points of 2S atoms with motional energy E. Because the turning points
satisfy e = 4rh2 ais-2s nEC rurn)/m+ U(rNu,(), this implies that the shape of the
spectrum is approximately given by

S(2hv) = 7Fh (0) d3r *N() NC()

47rr,2 aiS-2s nBEN r

x6 [2hv - (Es- 2 s+ 4 BEC 4() + U(r) - [IN

7rhQ2(0) 3 N
= d 3 r nBEC

4aisis) nEC r)\1

x6 2hv - Eis- 2s + 4,h 2 (als2S- m , (6.31)

which is identical to Eq. 6.17. In the last line we have used the fact that in the
Thomas-Fermi approximation, I-N - U(r) = 47rh 2 ais-is nNEC (r)/m.

Numerical calculation of the overlap integrals in Eq. 6.29 for a spherically sym-
metric harmonic trap confirms Eq. 6.31 (Fig. 6-5). The integral over the BEC density
distribution can also be motivated by noting that when a Lyman-alpha fluorescence
photon is detected at a given frequency in the spectrum of the condensate, it is
recording the fact that a IS atom was found at a position which had a iS density
which brought that atom into resonance with the laser. The rate of detection of such
photons is proportional to the probability of finding a condensate atom in a region

99
 1.2 1.2I I I I I

1.0

0.8

-50.6
C I

0.2

0 - .. -
-1.0 -0.8 -0.6 -0.4 -0.2 0
Laser Detuning [MHz at 243 nm]

Figure 6-5: Theoretical Doppler-free spectrum of a condensate at T = 0 in a three-

dimensional harmonic trap. The dashed curve (Eq. 6.32) follows from the integral over
the BEC density distribution, Eq. 6.31, using the Thomas-Fermi density distribution
for a peak condensate density of 5 x 1015 cm 3 , assuming ais-is = 0.0648 nm and
as-2s = -2.9 nm. The stick spectrum results from the sum over the transition
amplitudes expressed in Eq. 6.29 for the same conditions. In the overlap integrals,
wave functions for an infinite harmonic trap were used for the 2S motional states.
These deviate from the actual motional states near the top of the BEC interaction
well, so the deviation between the dashed curve and the quantum mechanical stick
spectrum nearer zero detuning is not surprising.

with the correct density.

For a Thomas-Fermi distribution (Sec. 6.2.1) in a three dimensional harmonic
trap, Eq. 6.31 reduces to

15,whQ2(0)N (E 1 s- 2s - 2hv) F 2hv - E 1 s- 2s 1/2

S(2hw)
842(vs2
8 [47rh(as-2S-as~s)
-
lBEC(O)]
2 L 4h 2
(a1s-2s-as-is) nBEC(O)

(6.32)

for 4wh2 (ais-2 s - ais1s) TBEC(0)/m < 2hv - EIs-2s < 0, and otherwise S(v) = 0.
The number of atoms in the condensate, N, is related to nBEC(0) through Eq. 6.21.
Note that this result is independent of the oscillation frequencies and is valid for
spherically symmetric, as well as asymmetric traps.
Theory and experimental data are compared in Fig. 6-6. The statistical error bars

100
 120

100 X 1/40

-U 80

8 60

20 -

0'
-1.0 -0.8 -0.6 -0.4 -0.2 0
laser detuning [MHz]

Figure 6-6: Doppler-free spectrum of a condensate: comparison of theory and ex-

periment. The narrow feature near zero detuning is the spectral contribution from
the noncondensed atoms. The broad feature is the spectrum of the condensate. The
dashed curve is the semiclassical expression for the spectrum using the Thomas-
Fermi density distribution for a harmonic trap (Eq. 6.32), a peak condensate density
of 1 x 1016 cm- 3 and the measured value of als-2s = -1.4 nm. The spectrum is
discussed in more detail in Chap. 7.

for the data are large due to the small number of counted photons, but the theoretical

BEC spectrum for a condensate at T = 0 describes the shape of the data reasonably

well. We have used the value for ais-2s found in Sec. 5.2.1 from the spectrum of a
3
noncondensed gas, and a peak condensate density of 1.0 ± 0.2 x 1016 cm- .

It is worth noting that the description of the BEC spectrum given by Eq. 6.29 is
quite general. If the 1S-2S interaction were repulsive, this would simply modify the

effective 2S Hamiltonian, Eq. 6.27, and the form of the motional states excited by

the laser would change. Raman spectroscopy has recently been used to excite atoms

out of alkali metal condensates [107, 108]. In the regime where mean field effects

dominate, the formalism presented here should explain the excitation spectrum for

this process as well.

101
6.3 Cold Collision Frequency Shift for an Arbi-
trary System
In an arbitrary geometry, the density distribution and the excitation may be inhomo-
geneous. This implies that at a given laser frequency, the probability that an atom
is excited may depend on its motional state (See, for example, Sec. 4.3.2.). Also, if
the laser is not spatially homogeneous, the motional state of an atom can change as
it is excited to the 2S state, as discussed in Sec. 4.4.5 in the context of coherence
sidebands on the spectrum. This complicates the derivation of an expression for the
shift which would be analogous to Eq. 6.14.

6.3.1 Sum Rule for the Mean Frequency Shift in the Spec-
trum

L. Levitov[109] has recently derived a sum rule which relates the mean frequency
shift, weighted by the spectrum intensity, to an average mean field shift in the sample,
weighted by the local density and excitation rate (Fig. 6-7). This is

f dv (2hv - Eis- 2s)S(v) 47h 2

(ais-2 s - ais-is) f d3 r I 2 (r)[n(r)]2G(r,0)
f dv S(v) m f d3 r 12 (r)n(r) . (6.33)

Here, S(v) is the signal at laser frequency v, 1(r) is the laser intensity profile, and
G(r,0) is the normalized density-density correlation function for zero separation at
position r in the inhomogeneous sample[100],

G(r, 0) =-r)((i-r) N;0 N; (r) (6.34)

For a noncondensed system in thermal equilibrium G(r, 0) = 2 and for a Bose-

condensed system at T = 0, G(r, 0) = 1. G(r, 0) is well defined for any arbitrary
thermodynamic state, even one which is not in thermal equilibrium, and in principle
G(r, 0) can vary with r.
For a homogeneous system, G(r, 0) = g( 2 )(0) and Eq. 6.33 reduces to Eq. 6.14. The

102
 1000
-1S-2S signal (a) -- Excitation profil (b)
-- laser detuning - BEC density
500 ~-- Normal density-

-500 - -77-. -
- --
- -
-1000
-900 -700 -500 -300 -100 100 -150 -100 -50 0 50 100 150
Laser Detuning [kHz at 243 nm] Radial Displacement [m]

4
M (d)

2 ... .... ...(C).

...... .

0 0
x__

-2

-4
-900 -700 -500 -300 -100 100 -150 -100 -50 0 50 100 150
Laser Detuning [kHz at 243 nm] Radial Displacement [m]

Figure 6-7: Graphical depiction of the sum rule. Plot (a) represents a typical Doppler-
free spectrum of a Bose condensate and the noncondensed atoms. (See Fig. 6-6 and
Chap. 7 for more discussion of the spectrum.) The mean frequency in the spectrum,
weighted by the spectral intensity, is proportional to an integral over all space of
the product of the square of the total density and the excitation probability (Eq.
6.33). These factors are shown in plot (b). Plots (c) and (d) display the appropriate
products which form the integrands for the numerators on each side of the sum rule
equation (Eq. 6.33), showing that the condensate dominates the expression.

sum rule agrees with the result that for the same density, the shift in a noncondensed
gas is twice the shift in a pure condensate.

Derivation of the Sum Rule

Equation 6.33 is a powerful relation for analyzing the spectrum, but what is the
physics it contains? We can derive this relationship in the context of an explicit
calculation of the spectrum for an arbitrary system, and this will give us some insight.
To calculate the spectrum, we consider the cold collision frequency shift and mo-
tional effects. By implicitly allowing the motional states of atoms to change during

103
the excitation, we include effects such as time-of-flight broadening or coherence side-
bands. We neglect the AC Stark shift, laser linewidth, and all relaxation processes
such as photoionization and radiative decay of the 2S state. For the collisional inter-
action, we use the Hamiltonian, H', given in Eq. 6.8. We know that the frequency
shift is dominated by the 1S-2S interaction, so to a good approximation, we can
set ais-is = 0. This will greatly simplify the analysis. For the state of the sys-
tem before excitation, we take a general state with N IS atoms, denoted by | 1 IN;O).

This state could describe bosons, fermions, or even classical particles [110], in any
thermodynamic state.
We saw in Sec. 6.1 that for weak excitation the transition frequency shift is in-
dependent of the number of 2S atoms, so, for simplicity, for the state after laser
excitation we consider only configurations of the system with 1 2S atom and N - 1
IS atoms. Note that these configurations form a manifold of states of the system.
In each configuration, the 2S excitation is associated with different single-particle
motional states. In the absence of H', this manifold is degenerate, and the energy is
about equal to EIs- 2s (neglecting kinetic energy). The interaction breaks this degen-
eracy, and one can then think of the laser exciting the system to one or a distribution
of the eigenstates of H', as shown in Fig. 6-8. We denote the eigenstates as ivi), and
H'|vi) = Eflvj).
The states and energies are not determined until we specify the trap geometry,
but we can write the expression for the spectrum,

S(2hv) = |(vijHiasIIN;o) 2 6(2hv - E 1 s- 2 s - E'). (6.35)

The overlap matrix element is calculated with the atom-laser Hamiltonian, which can
be written (Eq. 4.8)
N hQ R0 (I2)( Jj(-6
H as =_E 2 (2)(S).(.)
j=1 2
We can show that Eq. 6.35 implies the sum rule, Eq. 6.33. We start with

2h J dv (2hv - Ejs- 2s)S(2hv) =

104
 N-1 1S) 1 2S
E1S-2S
Iv.>

2hv

0 N 1S, 0 2S

Figure 6-8: Excitation of the system to excited state ivi). The states with energy
around Eis-2 s are the eigenstates of H' with 1 2S atom. The distribution of states
which can be excited by the laser leads to inhomogeneous broadening of the 1S-2S
spectrum.

= 2h dv (2hv - E 1 s- 2s) 2(ViHIas N;O 12 6(2hv - EIs-2s- Ej)

2
FE ;,|TN;OILtaI ) |\I
- L.. ~~"as IVi/ \Vi Ils s ITN;O

;HIasH'Z\vi (ViHas
7lN N;0

The eigenstates fvi) are a complete orthonormal basis for the Hilbert space connected
to 4pN;o) by Hias, so the expression further simplifies to

2, FN;|HitasH'Hias 4 N;O

47Th 2 ais-2 s J d3r 7thQ0(r) [n(r)] 2 G(r, 0).

(6.37)
m f 2

In the last line we have inserted the form of Hs and H' and identified the correlation
function G(r, 0). The details of this calculation are given in Sec. D.5.
Using the same techniques, one can show (Sec. D.5)

2h dv S(2hv) =

105
 = 2h dv (v1|Hlas IN;O) 2 6(2hv - E 1 s- 2 s - E )

2
7(4,N;O|HtasVi)(VIHlas N;0

=h( N|H Hiss|N;) d3 7 h (r)() (6.38)

Taking the ratio of Eq. 6.37 and 6.38 and recalling that Q0 (R) oc I(R) (Eq. 4.17)
proves Eq. 6.33 within the approximation that ais-is = 0. It is straightforward to
include ais-is 5 0 in the proof.

Discussion of the Sum Rule

The calculation in the previous section shows that Eq. 6.33 is reminiscent of other sum
rules in atomic physics, such as the Thomas-Reiche-Kuhn sum rule for the oscillator
strengths in electronic transitions in an atom [111, 112]. The physics in Eq. 6.33
is that if the degeneracy in a manifold of states is lifted by a two-body interaction
like H', then there is a relation between the mean frequency in the spectrum for
excitation into that manifold and a weighted spatial average of a local expression
for the interaction. In this case, we identify the local expression as AEc011(r) =
(4wth 2 ais-2s/m)G(r, 0)n(r), and the weighted average is

f d3 rQ2(r)n(r)AEc
1 1 (r)
I dar0
D(r)n(r) . (6.39)

It is important to note that the sum rule says nothing about the shape of the
excitation spectrum. In principle, the recipe implied in the expression for the spec-
trum given in Eq. 6.35 can be used to calculate the lineshape, but diagonalizing the
interaction Hamiltonian and finding the overlap integrals for a realistic system is a
daunting task. For a homogeneous system this can be done, which is essentially the
calculation in Sec. 6.1. In an inhomogeneous system, however, the motional state of
a particle can change when it is excited to the 2S state, and the number of configu-
rations which must be included in the interaction Hamiltonian becomes too large to
handle for even a few hundred particles.

106
Accuracy of the Semiclassical Calculation Used to Model Spectra and Ex-
perimentally Calibrate the Cold Collision Frequency Shift

The sum rule implies that the cold collision frequency shift probes the quantum nature
of the system. This is evident from the appearance of G(r,0). For a system in thermal
equilibrium with a given density profile and excitation profile, the mean frequency
shift in the spectrum for classical particles (G(r,0))=1) is one half of the mean shift
for noncondensed bosons (G(r,0))=2) 4 .
This calls into question the validity of the semiclassical calculation of the lineshape
(Sec. 4.4) which is used in Chap. 5 to analyze the data for the cold collision frequency
shift. For that calculation, a local shift of the level spacing,

AEc0 ii(r) = (8-Wh2 /M) (als-2s - ais-is)n(r), (6.40)

is included in the Hamiltonian which is used to numerically calculate the contribution

to the Doppler-free 1S-2S spectrum for an atom which moves in the trap. Particles
are treated classically in this calculation. It is not possible to include the effects
of bosonic symmetry which are essential for making (G(r,0))=2) and producing the
exchange contribution to the cold collision frequency shift (See Sec. 6.1.3).
It seems that setting AEcii(r) = (4-Fh 2 /m)(ais- 2s - ais-is)n(r) and using the
semiclassical numerical calculation of the lineshape accurately calculates the spec-
trum for a classical gas. The mean frequency shift will then be one half of the mean
frequency shift observed for a Bose gas with the same density distribution and ex-
citation geometry. In Chap. 5, we calculated the spectrum semiclassically, but used
,AEco(r) = (87rh2 /m)(ais- 2s - ais-is)n(r), so the mean of the calculated spectrum
should equal the mean of the spectrum for the experimentally observed Bose system.
The calibration of the cold collision frequency shift, and the extracted value of
ais-2s given in Chap. 5, utilizes the semiclassical spectrum calculation to account for
the inhomogeneous density distribution. It would seem to be better to use the sum

'Far from quantum degeneracy, the difference between the Bose distribution and Boltzmann
distribution is negligible.

107
rule to analyze the data and calibrate the shift, but, since the calibration is most
sensitive to the mean shift in the spectrum, the above arguments imply that both
approaches should give the same answer.
For a given density and laser profile, it is found that the mean frequency shift
in the calculated spectrum is systematically -5% larger than the value found from
the right hand side of the sum rule, Eq. 6.33. Allowing for the approximations used
in the numerical calculation (Sec. B.2), this agreement is satisfactory and seems to
confirm that the numerical calculation adequately reproduces the mean frequency
shift in the spectrum. The 5% difference is small compared to other uncertainties in
the calibration of the shift, and thus it is not worthwhile to reanalyze the data for
the shift using the sum rule.

Potential Applications of the Sum Rule

If spectra are recorded carefully so that the entire lineshape is scanned and the back-
ground count rate can be taken into account, it is straightforward to determine the
mean shift in the experimental spectrum. The sum rule can then be used to extract
quantitative information from the data even when the lineshape is not well understood

(see, for example, Sec. 7.5).

One must still know the laser and trap geometry to evaluate the required integrals
for the sum rule, but these calculations require little computational time compared
to the semiclassical calculation of the lineshape which has been used (Sec. 4.4). In
the future, the sum rule should be a useful analytic tool.

6.4 Conclusion
Presently, there is no accurate and computationally executable model for the shape
of the spectrum for a noncondensed Bose gas at very high density. There is no reason
to expect that the semiclassical calculation of the spectrum, with twice the classical
frequency shift, accurately reproduces the lineshape. For an accurate simulation, one
might need to perform a fully quantum mechanical calculation (Eq. 6.35). In the

108
regime where the lineshape is not dominated by the cold collision frequency shift,
good agreement is found between the data and the semiclassical calculation (Fig.
5-1), but simulations for spectra recorded at higher density have not been successful
(Fig. 5-4). Uncertainties in the trap and laser geometry make a detailed study of the
lineshape difficult at this time, but if these uncertainties can be reduced (Chap. 8),
such a study could prove interesting in the future.
Improved accuracy in the theoretical calculation of ais-2s and the experimental
measurement of the proportionality between the density and frequency shift could
give more insight into the shape of the spectrum for the noncondensed gas. Further
theoretical investigation of the cold collision frequency shift would also be beneficial.
There are of course other questions regarding the mean field treatment of the
frequency shift. Hydrogen atoms can receive a small momentum kick when excited
Doppler-free due to the spatial profile of the laser beam (Sec. 4-8). A recent calcula-
tion [109 indicates that this will add a small additional shift and a broadening to the
spectrum. Also, we have ignored collisional diffusion of the excited 2S atoms. This is
a reasonable first attempt, however, at a mean field theory of the effects of collisions
on the Doppler-free 1S-2S spectrum of trapped hydrogen.

109
Chapter 7

Spectroscopic Studies of a
Quantum Degenerate Hydrogen
Gas

An exciting new application of 1S-2S spectroscopy is in the study of Bose-Einstein

condensation. A general discussion of BEC in hydrogen is found in D. Fried's PhD
thesis [23]. Here, we emphasize the utility of 1S - 2S spectroscopy for studying
the condensate and point out some of the new physics probed with the two-photon
transition.
The appearance of two new features on the 1S-2S spectrum provides the clearest
indication that the BEC phase transition has been crossed. Figure 7-1 shows these
features, which, for the particular loading conditions and evaporative cooling path
used, appear when the sample is cooled below about 70 pK. These new features arise
from excitation of condensate atoms.

110
 500
A
400 5000 counts/s

4-0

o300 - Doppler-Free

C-)

200-'\\-
T Doppler-Sensitive
0
100--

-2 0 2 4 6 8 10 12

Laser Detuning [MHz at 243 nm]

Fiue7-1: Composite 1S-2S two-photon spectrum of trapped hydrogen. o-spectrum
of sample without a condensate; .- additional features due to presence of a condensate.
The intense, narrow peak arises from Doppler-free absorption of counter-propagating
photons by the normal gas. The Doppler-free contribution from the Bose-Einstein
condensate is shifted to the red and broadened due to atom-atom interactions in the
high density region. The wide, low feature on the right is from Doppler-sensitive ab-
sorption of co-propagating photons. The fact that the condensate contribution near
the center of the Doppler-sensitive line is narrower than the broad signal from the
normal gas confirms the condensation in momentum space expected with BEG, al-
though like the Doppler-free BEG signal, the Doppler-sensitive contribution is shifted
and broadened due to interactions.

7.1 Spectroscopic Signature of Bose-Einstein Con-

densation

Turning first to the broad Doppler-sensitive peak, recall (Sec. 4.2) that this feature
is shifted 6.7 MHz from the unperturbed 1S-2S frequency by momentum recoil, and

111
the lineshape maps the axial momentum distribution. The condensate feature near
the center of the line is narrower than the contribution from the normal fraction,
showing the condensation in momentum space. In the absence of interactions, the
width of this feature would be determined by the axial momentum distribution in
the condensate, which is approximately given by the position-momentum uncertainty
relation, Ap, ~ h/Az, where Az is the axial extent of the condensate. This would
correspond to a spectral width of less than a kHz. However, because of the cold
collision frequency shift, the BEC feature has a width of about 1 MHz and is shifted
from line center.
The Doppler-free contribution to the spectrum can be seen in Fig. 7-1 near v = 0.
The shoulder at large red-shift results from excitation of condensate atoms. It is
red-shifted from the intense narrow Doppler-free spectrum of the normal fraction by
the cold collision frequency shift, and is evidence of the high density which results
from condensation in position space, as expected in a trapped gas.

7.2 Doppler-Free Spectrum of the Condensate

Figure 6-6 shows the Doppler-free spectrum of the largest observed condensate. Sec-
tion 6.2 discusses in detail how the spectral lineshape can be related to the dis-
tribution of densities in the condensate. Because the cold collision frequency shift
dominates lineshape, the spectrum essentially is a histogram of density in the con-
densate. Using the value of als-2s found from spectroscopy of the normal gas (Sec.
5.2.1), and describing the condensate with the equilibrium system at T = 0 for which
2hAVB e 4Fh 2 (ais-2s - ais-is)nis/m (Sec. 6.2), the peak shift in the spectrum
implies that the peak condensate density is n = 1.0 ±0.2 x 1016 cm- 3 .
One should consider the possibility of extra broadening or shift of the resonance in
the BEC spectrum because the cold collision frequency shift was calibrated at densities
two orders of magnitude lower than those encountered in the condensate. Nonlinear
effects due to collisions between three or more particles should still be negligible even
for excitation of condensate atoms, because the parameter governing the importance

112
 3
of higher order collisions, a 1s-2sns3, is always small. (als-2sn = 0.03 for n -s

1016 cm- 3 .) One cannot rule out the existence of other higher order effects which are

not governed by this perturbation parameter, but as will be discussed below, there is
no sign of such an effect.

7.3 Condensate Fraction

7.3.1 Theoretical Limit of the Condensate Fraction in Hy-

drogen

For N particles in an infinite harmonic trap [113, 114], the critical temperature for

Bose condensation is T, = hD[N/1.202]i/ 3 , where W is the geometric mean of the

trap frequencies. At a temperature T < Tc, the fraction of atoms in the condensate,

neglecting interactions, is

f = NB/(NN + NB) 1 - (T/Tc) 3, (7.1)

where the subscripts B and N refer to BEC and normal, respectively. For Rb and Na

atoms, it is possible to evaporatively cool to T < Tc and achieve condensate fractions

as high as 75%[106] in a purely magnetic trap. Hydrogen behaves differently, however.

The group of Walraven et al.[115] calculated the dynamics of trapped hydrogen
at T < T, under conditions essentially identical to those in the experiments discussed

here. For an evaporatively cooled system in thermal equilibrium which crosses Tc

at a density of - 1014 cm- 3 , they concluded that the maximum condensate fraction

should not exceed a few percent.

Below Tc, the high density in the condensate, even for a low condensate fraction,
produces a very high dipolar loss rate from the condensate. The condensate is fed
through collisions in the normal gas, which are infrequent because of the small hy-

drogen scattering length. This implies a limit for the condensate fraction at which
condensate atoms are lost as fast as they can be replenished. One could say that

trying to evaporate from T = Tc to T < T, is thwarted in hydrogen because the time

113
scale for the evaporation, set by the low elastic collision rate, is longer than the time
scale for the destruction of the condensate due to dipolar decay. Essentially, T, drops
as fast as T drops, so T is always close to Tc. In Rb[116] and Na[117], the collision
cross section is more than 1000 times the hydrogen cross section, and the evaporation
time from T = T, to T < T, is much shorter than the decay time of the condensate.
It is also instructive to consider the balance between heating and evaporative cool-
ing in the sample, which determines the temperature. For the hydrogen experiment,
this has been investigated by D. Fried [23]. In an elastic collision between two normal
atoms which results in one atom entering the condensate, the condensate atom carries
no kinetic energy, while the normal atom takes away all the energy initially in the col-
lision. Thus a net flux of atoms from the normal gas into the condensate, for instance
when the condensate is forming and growing or holding constant in size against the
ever present dipolar decay, increases the average energy per atom in the normal gas.
This imposes a large heat load on the normal gas which must be balanced by evapo-
rative cooling. The finite evaporative cooling power limits the attainable temperature
for a given number of trapped atoms, and thus limits the condensate faction.
In the hydrogen experiment, the condensate fraction can be determined in two
independent ways. We will discuss each method separately.

7.3.2 Spectroscopic Determination of the Condensate Frac-

tion

One can determine the condensate fraction by comparing the areas under the normal
and BEC contributions to the Doppler-free spectrum. It is experimentally difficult
to record both features simultaneously, but the trapping and cooling is reproducible
enough on short time scales to compare spectra of identically prepared samples.
To account for different experimental conditions, one must consider that the signal,
S,(vi), is recorded at discrete frequency points, vi, with spacing Av, where x is either
B or N. The excitation time per point is Tx. The geometric overlap of the laser and
the atoms is accounted for by a parameter f3 . If the laser beam is well aligned, the

114
 50

10-
I I I * II I

-0--Normal Gas-
1--_BEC
.....- Loser 11 1/e--
--- 1/2 laser waist (w(z 2
-30--

-50-
-5 -3 1 1 3 5
Axial Distance [cm]

Figure 7-2: Typical laser and sample geometry at the BEC transition. The mag-
netic trap is described in Sec. 5.1.1, and the sample temperature is 45 AK. The
contour is plotted where the density of the normal gas is 1/e of its peak value.
The extent of the Thomas-Fermi density distribution for a condensate with 10'
atoms is shown. The excitation rate goes as the laser intensity squared, I2 (r) =-
[2p/7rW2(Z)]2 exp [-4r 2/W2(Z)], where P is the laser power, w(z) = wor1 + (Z/zo)2 is-
the characteristic radius of the laser beam, and zo = 7rwo/A is the divergence length
or Rayleigh length. The contour where I2 is 1/e of its peak value is plotted, along
with w(z)/2. Note the different scales for each axis. The sample's axial length is
about 400 times its diameter.

condensate is entirely within the laser focus and #B ~..1, but the laser focus is smaller

than the thermal cloud (Fig. 7-2). Through numerical calculation (Sec. D.5), ON
can be found. It is defined as the ratio of the integral of the signal using the actual

experimental geometry to the integral of the signal with an infinite laser radius. It can
be interpreted as the effective fraction of noncondensed atoms which are illuminated

by the laser.
Assuming that in the laser the excitation probability is the same on resonance
for a condensate and normal atom, the number of atoms in component x is propor-
tional to the normalized integral of the contribution of component x to the spectrum,

115
Ei ()AUz/(#272). The condensate fraction is given by

Ei SB(Vi)APB /( 3 BTB)
E>i SB(Vi)AVB/( 3 BTB) ± Ei SN(Vi)ZAVN/( 3
NTN)

We calculate f for the data shown in Fig. 7-3 and the results are given in Fig.
7-4. The time evolution of the observed condensate and normal gas will not be
identical in this data set because there is significant loss due to laser excitation when
on resonance with the normal gas (Sec. 4.3.2). Thus only the calculated condensate
fraction immediately after the end of the forced evaportion will be reliable.

Corrections for Laser and Sample Geometry, and Uncertainties in the

Measurement

The value of / 3N is sensitive to the laser beam and sample geometry. The time-of-
flight linewidth observed for low density spectra implies that the effective laser beam
waist was ~ 45 pm when this data was recorded (See Sec. 4.3.2.) This is close to
the beam waist for which the optical layout was designed, adding confidence that the
laser geometry is well understood. The axial overlap of the laser and atoms is still in
question, however, the signals from the condensate and normal gas are both sensitive
to this in similar fashion, reducing the sensitivity of the fraction measurement to this
factor.
Because the thermal cloud is larger than the laser beam radius, /N is also sensitive
to temperature. The most reliable measure of temperature in this regime is the
Doppler-sensitive spectrum. When the condensate is present, Bose statisics must
be taken into account to understand the shape of the Doppler-sensitive line [23],
and the data indicates that the average sample temperature is approximately 60 PK
while the condensate exists, but there are large error bars for this measurement and
it is unclear if the sample temperature is changing while the condensate is present.
Presently, the time required to make this measurement is too long to use it to monitor
the changing temperature while the condensate exists. The temperature diagnostics
must be improved in order to measure condensate parameters accurately (Chap. 8),

116
 Normal Fraction
0 4- 4. 4
3 0.35 s 3 1.05 s 3 1.75 s 3F
U) 2 2 I
U)I 1 -
0- P01. 0
c-J
U) -70 -50 -30 -10 10 -70 -50 -30 -10 10 -70 -50 -30 -10 10
Laser Detuning [kHz at 243 nm]

4: 4 4
3 2.45s II 3 3.15 s IF 3 4.55 s
2- 2 70 51 2 . .
1 .. 1 1
0
-70 -50 -30 -10 10 -70 -50 -30 -10 10 -70 -50 -30 -10 10

4' 4
3 5.95 s 3 7.35s I 3 9.45 s
2 2 I 2 .
1 0 . 1 . .1 1 0 1
0 -70 . ..- , 1 0-.0
.,..-L_ . . 0
-70 -50 -30 -10 10 -70 -50 -30 -10 10 -70 -50 -30 -10 10

Condensate
60 60 60
50 0.3 - 50 0.9 s 50 1.5 s
C
40 40 40
30 30 30
20 20 20
U) 10 10
10
Cn- 0 0 0
-1000 -750 -500 -250 0 -1000 -750 -500 -250 0 -1000 -750 -500 -250 0
Laser Detuning [kHz at 243 nm]

60 60 60
50 2.1 s 50 2.7 s 50 3.3 s
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
-1000 -750 -500 -250 0 -1000 -750 -500 -250 0 -1000 -750 -500 -250 0

60 60 60
50 3.9 s 50 5.1 s 50 7.1 s
40 40 40
30 30 30
20 20 20
10 10 10
0 .,.00 ,.. , , , 0 10. ... ,
2. 0
-1000 -750 -500 -250 0 -1000 -750 -500 -250 0 -1000 -750 -500 -250 0

Figure 7-3: Doppler-free spectra of normal gas (each data point represents 50 ms of
excitation) and condensate (each data point represents 300 ms of excitation). Sam-
ples are identically prepared and the trap depth at the end of the forced evaporation
is 280 pK. The time quoted is the time of the midpoint of the scan, measued from
the end of the forced evaporation. The laser power in the trapping region is about
5 mW. The effective beam waist is 45 pm, and is determined from the time-of-flight
linewidths of the later, low density spectra of the normal gas.

117
 1000 0.04 . . .
(a) (b)
__ 800 o Noncondensed Gas .0 0.033 - T=50 .K
* Condensate (xl0) T=70 AK
~J600 U.

480 0.02

0 0.01 - -
200 -oi1:

0 0'
0 5 10 15 20 25 30 0 1 2 3 4 5 6 7
Time [s] Time [s]

Figure 7-4: Spectroscopic determination of the condensate fraction. (a) Total counts
in one scan for the data shown in Fig. 7-3. The quoted time for each sweep is the
mean time after the end of the forced evaporation. Error bars are statistical, but
additional uncertainties due to variations in laser power and alignment are indicated
by the spread in the data from smooth curves. As discussed in the text, only the initial
condensate and normal gas spectra can be compared directly because the loss rate due
to excitation is significant when on resonance with the normal gas. (b) Condensate
fraction determined as described in the text. To calculate the fraction, one must know
the sample temperature. The temperature is expected to be about 60 pK. The error
bars represent statistical uncertainty only. Only the t = 0 measurement is reliable. If
data could be taken in such a way that the time evolution of the sample is identical
when recording both the condensate and normal gas spectra, than the time evolution
of the condensate fraction could be measured in this way.

and the time dependence of the sample temperature in the presence of the condensate

is not yet well understood.

Assuming thermal equilibrium of the normal gas and using a nontruncated bose

particle distribution, N = 1/8 for T = 50 pK and N = 1/12 for T = 70 pK. Using

the actual truncated distribution in which no atoms with energy above the trap
threshold are present [23], the factors are ~ 1/7 for T = 50 pK and N 1/8
for T = 70 pK. Using the latter values, the condensate fraction immediately after
the end of the forced evaporation, calculated from the ratio of the spectral areas,

is f 3%,
+ where the errors are chiefly due to uncertainties in the experimental

geometry and sample temperature. The larger positive error reflects the difficulty in

ensuring that the laser is perfectly aligned over the condensate. Experimentally, it

118
is seen that slight misalignment can decrease the condensate signal by as much as
50%. The number of atoms in the condensate implied by this measurement is very
sensitive to the temperature of the sample, but a conservative estimate is greater than
109 atoms.

7.3.3 Determination of the Condensate Fraction from the

Peak Shift in the Doppler-Free Spectrum

If the condensate is well modelled by an equilibrium system at T = 0, then the density

distribution is given by the square of the Thomas-Fermi condensate wave function
(See Sec. 6.2.1), and the number of atoms in the condensate is given by

29 / 2 7(U 3/ 2 (nBEC)5/2,
N = p(7.3)
15m 3/2wwz

where U 4wh2 ais-is/m. Using calculated and measured trap parameters, a peak
condensate density of n E = 1.0 ± 0.2 x 1016 cm- 3 corresponds to 6 ±3 x 109

condensate atoms, where the uncertainty reflects only uncertainty in n BC, not in
the trap parameters.
If the system is in thermal equilibrium, the number of atoms in the normal phase,
NN, is determined if the temperature and trap are known and T < Tc. NN varies
3 [23], and as discussed above, the temperature is difficult to
approximately as T-
determine. Assuming thermal equilibrium of the entire sample at a temperature of
about 60 pK, this condensate density corresponds to about a 9% condensate fraction
[23] with error bars of at least a factor of 2 arising from uncertainty in temperature
and nEi.

7.3.4 Implications of the Measurement of the Condensate

Fraction

The spectroscopic determination of the condensate fraction is independent of how one

models the condensate. Thus the reasonable agreement between the two methods is

119
a check that the equilibrium T = 0 model describes the BEC wavefunction well.
This should only be viewed as a preliminary result, however, because there are such
large uncertainties for these measurements. The dominant errors in the parameters
which determine the condensate fraction, either from the ratio of spectral areas or
the peak shift of the condensate spectrum, are systematic, not statistical. They
arise from imprecise knowledge of the sample temperature and the geometry of the
sample and laser beam. It is difficult to assign uncertainties to these parameters and
large systematic errors would obviously confuse the analysis. Improvements in the
experimental apparatus (Chap. 8) are required to minimize the possibilities for large
errors.

Evidence for Spatial Correlation in the Condensate

One can interpret the reasonable agreement between the two calculations as the first
evidence of local second order spatial coherence in the hydrogen condensate because
the equilibrium model assumes g( 2) (0) = 1. If coherence is lacking, g( 2) (0) = 2 and the
shape of the spectrum and the relation between the frequency shift and condensate
density would be different (See Sec. 6.1.3).
The study of phase coherence and spatial correlation in condensates in dilute gases
is a subject which has inspired much debate (See the discussion in [118]). A brief
review of the experimental evidence for coherence and correlation in alkali metal con-
densates is found in [101]. Long range first order spatial correlation (g(l)(r) = 1) in a
Na condensate was established by the observation of interference fringes in the spatial
distribution of atoms in two overlapping condensates[118]. The most direct measure-
ment of local second order spatial correlation (g(2) (0) - 1) comes from observations of
the ballistic expansion of Na[119, 101] and Rb[120, 114] condensates. Such data de-
termines the condensate's interaction energy, which depends on the s-wave scattering
length, density, and g(2 )(0). One could also take the agreement between theories and
experiments on collective excitations[121, 122], sound propagation [123], component
separation in mixtures of condensates[106, 124], and the limited condensate number
in the negative scattering length Li BEC [22], as evidence that g( 2)(0) = 1. All of these

120
 Time is mean time following rf evaporation. (0.84 s per sweep)
I-
60
0.4sec 2.1sec 3.8sec

40

E + normal gas
* condensate
20

+9
0-
u 60 -
5.5sec 7.1sec 8.8sec

S 40 -

20 -

0-
-800 -600 -400 -200 0 -800 -600 -400 -200 0 -800 -600 -400 -200 0
detuning [kHz] detuning [kHz] detuning [kHz]

Figure 7-5: Time evolution of the Doppler-free spectrum of a single condensate. The
trap depth is held at 280 pK. The time quoted is the time of the midpoint of the scan,
measured from the end of the forced evaporation. The condensate appears stable in
peak density and atom number for 2-3 seconds, and then decays away over the next
few seconds.

effects probe the interactions between condensate particles, and thus probe g(2 )(0),
and the theoretical models all assume an equilibrium condensate with local second
order spatial correlation. Local third order spatial correlation (g( 3) (0) = 1) in a Rb
condensate was demonstrated by comparing the three-body decay rates in condensed
and noncondensed samples[125].
Because spatial correlation is the rule in Na and Rb condensates it would be sur-
prising if second order spatial correlation were absent in H, but there are differences
which make its existence nontrivial. Dipolar decay from a H condensate with 10'
atoms causes the condensate to decay with a time constant of about 1 second. The
observed lifetime of about 3 seconds (Fig. 7-5) implies that the condensate is con-

121
stantly being replenished by new atoms from the thermal cloud, so coherence must
be reestablished continually. The loss rate in the H condensate is much higher than
that observed in the Rb and Na condensates because the density in the H condensate
is more than an order of magnitude greater.
One might suspect that coherence is established slowly because of the low hydrogen
elastic collision rate. (As pointed out before, the elastic collision cross section is 1000
times smaller in H than in Na and Rb.) The time scale for formationof the condensate
will be governed by this rate [126, 127], but forming a condensate and establishing
coherence or correlation are two different processes. It has been suggested[127] that
once the condensate forms, the establishment of coherence is fast, occurring on a time
scale of Tcoherence - h/(47rh2as-2sn BEC/m). This is on the order of milliseconds for
hydrogen, which implies that coherence is established quickly on the time scale of H
loss processes.

7.4 Phase Diagram

Because the normal gas and the condensate contributions to the Doppler-free spec-

trum are well resolved, one can study the hydrogen phase diagram near the BEC

phase transition (Fig. 7-6). The shift of the resonance in the normal gas is used to

measure the density through an analysis similar to that in Sec. 5.1.2. The data shows

that as the sample is evaporatively cooled, the atoms settle into the trap and the peak

density increases until the BEC phase transition is reached. At lower temperatures,

the normal gas density is pinned at nc(T). The critical density limits the population
in the normal gas and atoms which cannot be accommodated form the condensate.

Figure 7-6 shows that the largest condensate and condensate density is found

shortly after the transition is crossed. This is very different from the behavior observed

in Rb and Na condensates (see, for example, [114]), for which the condensate size

continues to increase as the sample is cooled. In hydrogen, the fast dipolar decay and

slow evaporative cooling, as discussed above in the context of the condensate fraction,

precludes penetration far into the BEC regime.

122
 10116
5

E 2
>10 0 Condensate
Fn 0 Normal Gas
a>
2 ..
E 14 .. .*
10,
5*

2
102 2 3 4 5 7 10 3
Trap Depth [puK]
Figure 7-6: BEC phase diagram of hydrogen and a typical evaporative cooling path.
Each data point represents a different load of the trap. The solid line is the BEC phase
transition, nc(T) = 2.612(27rmkBT) 3 / 2 /h 3 , assuming the sample temperature is one
sixth of the trap depth. The density of the condensate and normal fraction is measured
with the Doppler-free spectrum and the phase separation is clearly signalled by the
appearance of the shoulder at large red-shift as described in Fig. 7-1. The scatter of
the data for the normal fraction reflects the typical uncertainty in the measurement.

7.5 Spectrum of the Normal Fraction in the De-

generate Regime
Inspection of the Doppler-free spectra of the normal gas which were used to construct
Fig. 7-6 reveals some surprising behavior (Fig. 7-7). When the BEC threshold is
crossed, the spectrum qualitatively changes shape and has contributions at large red-
shift. Some shape change is expected when the chemical potential approaches zero
and the shape of the Bose density distribution ceases to be well approximated by the
Maxwell-Boltzmann distribution. However, much of the integrated signal intensity is
further red-shifted than the shift corresponding to the critical density. The volume

123
 1500
Temperature is approximately 1/7 of trap depth
Onset of BEC at trap depth =300gK
~1 [rap depth:
- 1070 pK
- 560 pK
- 430 gK
1000 - 290 K
above threshold
- 210 rK

500 below threshold

0-100 -50 0
frequency detuning [kHz]

Figure 7-7: Doppler-free spectrum of the normal fraction immediately after the end
of the forced evaporation. The signal in the presence of a condensate (trap depth
< 300 MK) stretches to larger red-shift than expected for a static system in thermal
equilibrium.

of the condensate is about 10-3 of the volume of the thermal cloud, so it is surprising
that the condensate appears to affect the spectrum of the normal gas so much.
In order to quantitatively analyze the spectrum, one would need improved temper-
ature diagnostics, knowledge of the trap and laser geometry, and theoretical under-
standing of the lineshape for the normal gas. However, the strange shape may imply
that the sample is far from equilibrium near the transition, so the critical density may
not be well defined. The sample's thermodynamic properties are changing rapidly in
the first few seconds after crossing the BEC threshold, as discussed in the context of
Fig. 7-4, and the large aspect ratio of the trap and the fast dipolar loss rates could
impede thermalization. Since the sample is always close to the critical temperature,
there may also be large fluctuations in the number of atoms in the condensate [128].
The data indicates that the degenerate hydrogen gas is a dynamic system and fur-
ther study is required to understand its kinetic properties and explain the interesting

124
shape of the Doppler-free spectrum of the normal gas.

7.6 1S-2S Spectroscopy as a Probe of BEC

High resolution spectroscopy is a powerful new way to study Bose condensation. It
already has been used to detect the condensate and study its momentum and density
distributions and its coherence properties. The frequency shifts provide information
on how impurity 2S atoms interact with the condensate and there is much to explore
in this area, such as collisional de-excitation. The spectrum should help to understand
the dynamic nature of the hydrogen condensate, such as the dipolar decay and the
feeding of the condensate from the thermal cloud.

125
Chapter 8

Future Prospects

1S-2S spectroscopy of trapped atomic hydrogen has progressed dramatically since

the signal was first observed in 1995 [10]. It is now a valuable probe of sample
temperature and density and has been used to study atom-atom interactions and
BEC. The technology is still evolving rapidly however, and there is promise for major
improvements.
The weak signal strength is presently a limitation when studying the condensate
or the Doppler-sensitive excitation of the normal gas. Shot noise due to the small
number of photons counted is the major source of error for these features: nearly a
second is required to record a useful spectrum. This makes study of dynamics and
thermalization very difficult.
A factor of 10 improvement in Lyman-alpha detection efficiency could be realized
by moving the atoms closer to the MCP detector to improve the solid angle for col-
lection of the fluorescence. Another factor of 6 improvement is available by replacing
the Lyman-alpha filter above the MCP with a MgF 2 window. As discussed in Sec.
3.4, the filter blocks scattered laser photons so that the MCP is not saturated. If the
filter were removed, to avoid MCP saturation the MCP gain could be reduced when
the laser is present, and restored on a time scale shorter than a millisecond in order
to detect the signal. Experiments have shown that switching the high voltage on the
MCP assembly can change the gain on a fast enough time scale. A possible concern
is that long-lived fluorescence from materials excited by the laser scatter (Sec. 3.3)

126
may produce a large background count rate if the filter is removed.
With greater signal to noise it should be possible to study the thermal equili-
bration of the sample near the BEC phase transition and to quantitatively compare
the condensate spectrum with the spectrum predicted by the Thomas-Fermi density
distribution. This should help to resolve the questions surrounding the strange ap-
pearance of the Doppler-free spectrum of the normal fraction near the transition and
improve understanding of transport properties in the quantum gas. Deviations from
the Thomas-Fermi prediction may provide imformation on finite temperature effects,
for instance.
In addition, a new trap geometry in which the atoms are moved away from the
large superconducting magnets before performing spectroscopy, could greatly reduce
the uncertainty in the trap and laser geometry. These uncertainties are the limit-
ing factors in the measurement of the cold collision frequency shift and the 1S-2S
scattering length, and in using the frequency shift to study the density distribution
of the sample. With better understanding of the trap shape it might be possible to
quantitatively compare the dipolar decay of a condensed and noncondensed gas and
directly measure the second order spatial correlation function, g (2 )(0).
While the laser illuminates the gas, helium atoms evaporate off the retroreflecting
mirror at the bottom of the cell and knock atoms out of the trap. This causes the
sample to decay with a - 10 second time constant. One would like to observe the
sample for longer and in a less destructive fashion. A helium film pump in the cell
[23] can be used to remove most of the liquid 4 He from the trapping region and reduce
the helium flux. This tool has not yet been used to its full potential, but preliminary
results in a previous experiment indicate that it increases the sample lifetime by a
factor of three to five.
Stray electric fields in the plastic cell used in the experiments described here limit
the lifetime of the excited 2S atoms to a few milliseconds, but in a metal cell the
lifetime approached the 122 ms natural lifetime [10]. It should be possible to coat
the inside of the plastic cell with a thin conductive layer to provide DC electric field
shielding without causing too much eddy current heating during the RF evaporation.

127
This will allow studies of the 2S lifetime in the condensate which may give more
insight into the interactions of these impurity atoms with the condensate.
A long 2S lifetime is essential for higher resolution spectroscopy of the 1S-2S
transition. Straightforward application of available technology in laser frequency
stabilization should improve the laser linewidth to 100 Hz or less. The production of
cold atoms which suffer less than 10 Hz residual Zeeman broadening is now feasible
with RF evaporation and adiabatic expansion of the cloud. With improved detection
efficiency, there would be no prohibition to working at densities below 1010 cm- 3
where the cold collision frequency shift becomes unimportant. The path seems clear
to achieving a spectral resolution at least an order of magnitude better than the best
results from atomic beam experiments. This use of the 1S-2S transition in trapped
hydrogen as a frequency standard seems promising.
Another avenue which has not been explored is the excitation of transitions from
the 2S state. When excited Doppler-free, the 2S atoms are trapped and are at ap-
proximately the same temperature as the IS atoms. Powerful techniques for imaging
in the visible wavelength region have been developed to observe small numbers of
alkali metal condensate atoms. In a 1 ms laser pulse, up to 106 2S atoms can be
excited. This number of atoms can easily be detected with absorptive or dispersive
imaging on the 656 nm Balmer-alpha transition, for example. This could prove to
be a useful alternative method for monitoring the 1S-2S excitation rate, but it also
could lead to new experimental measurements, such as the determination of numer-
ous scattering lengths for collisions between IS atoms and more highly excited states
than n = 2.

This is just a short list of the many improvements and experiments for 1S-2S
spectroscopy of trapped atomic hydrogen. The results are already impressive, but
they promise to be even more so in the near future.

128
Appendix A

High Resolution Spectroscopy

The two-photon 1S-2S transition in a cold, trapped sample has great potential for
high resolution spectroscopy. This goal has not yet been aggressively pursued at MIT
because achievement of BEC has been the primary objective. Even with the present
apparatus, however, the resolution is comparable to the best results from atomic beam
experiments[64, 9]. 1S-2S spectroscopy in an atomic beam is limited by the finite
interaction time of atoms with the laser, and by the second order Doppler shift of
the frequency. A trap provides long interaction times and a low enough temperature
that the second order Doppler shift is negligible.

A.1 Best Achieved Resolution

A.1.1 Time-of-Flight Broadened Lines

Figure A-1 shows successive scans of a cold (<40 tK), low density (<1013 cm- 3 )
hydrogen sample. A simple model for the spectrum incorporating the time-of-flight
lineshape and some additional broadening implies that the spectrum is still dominated
by the time-of-flight shape. Additional broadening, most likely arising from the laser
frequency spectrum, cold collision frequency shift, and photoionization, contributes
about 1 kHz FWHM to the width. The relative importance of each broadening
mechanism was not determined.

129
 on
120 $'FWHM=. k z 9 W=3.3 kHz kH
100
C. 80 60 60
~60
40 30 30

20

0 "rI- 0~ ~~ II '"I, 0 L' .I " 0 A-

-15-10 -5 0 5 10 15 -15-10 -5 0 5 10 15 -15-10 -5 0 5 10 15
Laser Detuning [kHz at 243 nm]
70 =28kz70
70 FWHM=2.8 kHz 60 . FwHM=2.8 kHz 70 FWHM=2.8 kHz
60 T 60-
50 50 50
40 40 40
30 30 30
20 20 20
T T
10 10 10
0 OU
' "* i 'i 0 "Ila
0 0 1'I
. 0 "U, i
-15-10 -5 0 5 10 15 -15-10 -5 0 5 10 15 -15-10 -5 0 5 10 15

Figure A-1: Spectroscopy of a single cold (<40 piK) low density (<1013 cm- 3 ) hy-
drogen sample. The sample was created by evaporating to a temperature of about
150 piK and then adiabatically expanding the magnetic trap. Each data point repre-
sents 25 ms of laser excitation, and each successive sweep was recorded in 5 s. The
excitation laser power was about 5 mW. The error bars are statistical and the solid
lines are the results of a simple model which incorporates the time-of-flight lineshape
and additional broadening mechanisms. The resulting linewidths are about 3 kHz
FWHM.

On the time scale of about one second, the laser linewidth is below 1 kHz, but on

a longer time scale, the laser shows more instability. For instance, between the third

and fourth sweep the spectrum shifts about 1.5 kHz.

For spectroscopy of trapped hydrogen, the signal rate is quite high. It is currently
possible to determine the line center to better than 100 Hz in one second. The high
signal rate and signal-noise ratio for trapped hydrogen is an important consideration

when comparing its potential as a frequency standard with that of trapped ions or

the low velocity tail of a hydrogen atomic beam.

130
A.1.2 Coherence Effects

Time-of-flight broadening dominates the linewidths in Fig. A-1. As discussed in

Sec. 4.4.5, in a tightly confined trap the frequency for periodic atomic motion is high
enough that atoms maintain coherence with the laser as they repeatedly pass through
the excitation region. This produces interference fringes on the spectrum which are
narrower than the time-of-flight linewidth [10, 60]. The envelope of the line is given
by the time-of-flight lineshape and the central fringe arises mostly from excitation of
very low energy atoms which are in the laser for a long time.
Presently, the width of the fringes is limited by the resolution limit of the appa-
ratus, - 1 kHz under ideal conditions. Figures 4-8 and A-2 show examples of spectra
exhibiting motional sidebands.

A.2 Laser Frequency Stability Limitations

By using the 1S-2S signal as our absolute frequency reference we were able to study
many systematics which affect the frequency stability of our laser system. Two im-
portant effects are discussed below. This section is intended to be useful for future
operators of the MIT laser system, and a familiarity with the system is assumed.

A.2.1 Reference Cavity Shift with Light Power

The frequency of the reference cavity transmission modes was found to vary with the
circulating light power in the reference cavity. The circulating power is monitored by
measuring the transmitted light after it passes through a 25/75 beam splitter. The
weaker beam falls on a DT-25 photodiode with a 1 kQ load which is amplified by 50
in the frequency servo electronics box. This signal is used to determine if the laser
frequency is locked to the cavity mode, which is important for the redundant cavity
locking scheme. Including the amplification, the calibration is about 0.4 volts per 100
[pW of cavity transmission. The stronger beam falls on an FND-100Q photodiode
which has a 22 kQ load and is amplified by 10 by a simple op-amp circuit. The

131
 180 350
300
0
120 250
2.3 kHz 200

C6 - 4 - 150
u), 60
100
50
0 0
-15 -10 -5 0 5 10 15 -15 -10 -5 0 5 10 15
Laser Detuning [kHz at 243 nm]
150
600
125
500 1.0 kHz
100
1.6 kHz 400
75 - -4
300
50
200
25 100
0 0 I I I -
I I I I I|

-10 -5 0 5 10 -15 -10 -5 0 5 10 15

Figure A-2: Spectra showing motional sidebands. Conditions vary, but typically,
each point represents 25 ms of laser excitation and a single laser pulse was 430 Ps
or greater. The width of the central fringe is indicated in each plot, and under ideal
conditions the best resolution is about 1 kHz, which corresponds to 1 part in 10-12.
Error bars are statistical and do not include uncertainty arising from variation in
laser power and alignment. The solid lines are meant to guide the eye.

calibration for this signal is 4 volts per 100 pW transmission.

We see 100 piW transmission for about 4 mW of incident power when each EOM
sideband has 30% of the power of the carrier. When the cavity was constructed [59],
the mirror transmission was measured to be about 0.1%. This implies about 100 mW
circulating in the cavity, which translates into an enhancement in the cavity of 50,
assuming about one half of the incident power is coupled into the cavity. In the
absence of any scattering or absorption at the mirrors, the enhancement would be
~ 640, where F is the cavity finesse. A 0.3% absorption or scatter, which is not

132
 I I I'

80

Q)
60 -

C)
20
0

60 70 80 90 100 110 120

Power Transmitted Through Reference Cavity [pW]

Figure A-3: Reference cavity transmitted power as measured by the FND-100Q pho-
to diode/ amplifier circuit. The relative frequency is found by comparing the mode
frequency to the 1S-2S transition frequency. The line is a linear fit which yields a -1.3
kHz/pW shift of the mode frequency. Long term drift of about 8 kHz/hour is taken
into account.

unreasonable, would reduce the enhancement by an order of magnitude[129].

Figure A-3 shows the sensitivity of the cavity mode frequency to circulating power.
We found that the mode frequency shifts by -1.3 kHz per pW change in transmitted

power. The data was taken over the course of a few hours and for each data point the
power in the reference cavity was changed approximately 10 minutes before the data
was taken. Our interpretation of this shift is that there are local heating effects on
the mirror coatings which change the effective length of the cavity. The time constant
for this effect is on the order of 100 ms, as shown in figure A-4.
The light level in the cavity varies due to laser power fluctuations and angular

displacements of the laser beam which change the coupling efficiency. If not controlled,
this would significantly broaden the laser spectrum. For the majority of data discussed
in this thesis, the power incident on the reference cavity was stabilized by varying
the RF power on an acousto-optic modulator (AOM). This reduced much of the

frequency shift, but even with an optical fiber between the laser and the reference
cavity, there were still slow drifts of the beam alignment which led to variations of

133
 350 103

E 300
0
102 D
C
(D 250

200 101
En200--

150 0
100 3

CN
99
2 50 -

0 98
0 1 2 3 4
Time [seconds]

Figure A-4: Response of the reference cavity mode frequency to a sudden change in
light level. The frequency offset of the laser from the reference cavity transmission
mode was held fixed and the 1S-2S excitation rate was monitored as the light power
in the cavity was changed. Each data point represents about 30 ms of time with 50%
duty cycle laser excitation. 250 counts per point represents the peak count rate on
line center. The lower count rate corresponds to the wing of the line about 5 kHz from
resonance. The 1S-2S excitation rate, and thus the cavity mode frequency, responds
to the change in light power with about a 100 ms time constant.

circulating power. Stabilizing the transmitted power is superior, and we now follow
this procedure.
We also see a shift of the mode frequency with a time constant of hours or more
if the light level is changed or when light is first put into the reference cavity, which
we interpret as a temperature change of the zerodur cavity mirror spacer. This effect
can be larger than the long term drift of the cavity of about -3 MHz per month.

A.2.2 Doppler-Shifts Along the Beam Path

As shown in Fig. 3-4, the laser system is about 30 m away from the hydrogen trap.
There are advantages and disadvantages to this arrangement. The dilution refriger-
ator generates significant noise and vibrations, which are not conducive to frequency
stabilization of a laser, so it is helpful to move the laser away. The laser beam must
still journey to the trapping apparatus, however, and at present no optical fibers

134
 FND GPD GPD
100Q 201 201 Mixer

+50 dB r C coupled FM
from MHz 146 MHz 8642B
laser notch

freq synth
73 MHz

30 meters
AOM
atoms

Figure A-5: Schematic of the Doppler servo system. The heavy black line shows the
laser beam path, which is in an interferometer configuration to generate a beatnote
at 146 MHz on the FND-100Q photodiode. The beatnote is amplified (GPD-201 is a
+30 dBm RF amplifier) and then mixed with a clean 146 MHz signal to generate an
error signal. The loop filter is a simple op-amp circuit with a 10 kQ input impedance,
gain of 1420 at DC, and a single pole at 8.4 Hz.

exist for 243 nm. Thus the beam bounces off mirrors mounted to the ceiling and,
eventually, the vibrating cryostat.
When a laser beam bounces off a moving mirror, its frequency is Doppler-shifted
by about 2k -v, where k is the wave vector of the light and v is the velocity of the
mirror. To investigate the size of this effect in the present experiment, the spectrum
of the light after it has traveled to the cryostat and returned to the optics table is
measured by interfering it with a beam which has not left the table (A-5). To move
the beatnote away from DC, an AOM frequency shifter is placed in the beam path.
Since the coherence time of the laser beam on the optics table, equal to the inverse of
the linewidth, is long compared to the travel time, the beatnote measures the spectral
broadening suffered by the laser beam along its path.
The beatnote after the mixer and its Fourier transform are shown in Fig. A-6. The
width of the spectrum is about 10 kHz. Since the laser is in a standing wave geometry,
and the atomic sample is essentially at the retro-reflecting mirror, the width of the
laser when it excites the atoms is about half of the width measured by the beatnote,

135
Tek stop 50.OkS/s 16 Acqs

Math2 Zooin: 2 OX Vert 0.2X Horz A: 10.00kHz

- : 10.00kHz

1-.

....
. ......

Ch1 200mV M2.50ms Chl X -96mV 9 Apr 1998

Iat12 10.0 dB 2.SOkHz 17:17:07

Figure A-6: Beatnote of Doppler servo system. The dark trace shows the beatnote
after the mixer. The scales are 200 mV/div and 2.5 ms/div. The light trace is the
fast Fourier transform, after subtracting the 146 MHz AOM shift, of the difference
frequency between the return laser beam and a laser beam which has not left the
optics table. The scales are 10 dB/div and 2.5 kHz/div. The implied FWHM of the
return beam laser spectrum is about 10 kHz.

and Doppler-jitter should limit the resolution of the experiment.

A control loop was built to vary the frequency of the AOM so as to lock the

beatnote to a radio frequency synthesizer. The schematic is shown in Fig. A-5. The
loop locks well with a 0 dBm beatnote before the mixer, and it acquires in less than
50 ps when the laser is chopped at about 1 kHz. When the loop is locked, the beatnote
width is less than 3 Hz. Since the laser beam passes through the AOM twice, to a
good approximation this scheme also produces a clean laser frequency spectrum in
the excitation region. Any jitter introduced by the retro-mirror will not be corrected.
Most of the data in Fig. A-1 and A-2 was recorded with the Doppler servo. A
systematic study was not done to determine the actual resolution limitation due to

136
the Doppler-jitter, or how much the Doppler servo helped. In most of the studies
described in this thesis the linewidth is greater than 4-5 kHz and can be accounted
for by the time-of-flight broadening and the cold collision frequency shift, and the
trap oscillation frequency is not high enough to resolve the sidebands. From the
width of the spectrum in Fig. A-6, it is still surprising that the Doppler-jitter limit
is not more clearly evident. In fact, on rare occasions the spectrum has very high
resolution without the Doppler servo, such as in Fig. 4-8 and the lower right hand plot
in Fig. A-2. Clearly, when high resolution spectroscopy is pursued more vigorously,
this effect will require further study.

A.3 Prospects for Improving the Frequency Sta-

bility
At present, the resolution of the spectrum is limited by the laser linewidth (- 1 kHz
at 243 nm), the low signal rate, and the cold collision frequency shift (One needs to
work at relatively high densities to obtain spectra with sufficient signal/noise). Laser
stabilization to Fabry-Perot optical resonators has progressed dramatically since the
current laser system was constructed [59]. A 1 Hz laser linewidth for an integration
time of over a minute has been achieved by J. Bergquist at NIST, Colorado [130], and
cryogenic optical resonators [131] are a recent advance which shows great promise.
As discussed in Chap. 8, the signal rate can be increased dramatically by improving
the detection efficiency for fluorescence - making it feasible to record spectra at much
lower densities.

137
Appendix B

1S-2S Spectroscopy Appendix

This appendix gives a more detailed derivation of the effective two-level Hamiltonian
and discusses the numerical simulation of the spectrum.

B.1 Effective Two-Level Hamiltonian

The present treatment of the photo-excitation spectrum begins with time-dependent
perturbation theory[78], with the perturbing Hamiltonian

' = - E(R, i, t). (B.1)

The operators are R, the center of mass of the atom, and i, the position of the
electron with respect to the nucleus. The charge of the electron is e < 0, and we
represent our standing wave laser as a classical monochromatic electric field of the
the form

E(R, i, ft= ±1E+(()eiki'(fi+F)-iwt+ % 2 E 2 (R)eik2-(Ri)-iW2t + c.c. (B.2)

E1 and E2 are taken to be real and they contain the slow spatial variations of the
beam profiles. The laser is applied at time to.
For an electric dipole-allowed transition, the excitation rate is found from the first

138
order coefficient of the excited state,

Cp2(t) = -- I dt"(2S, P2 I H'(t") iS, pl)e-i2s,2-s1(tto). (B.3)

As is well known, (2S, P2 I H'(t") I IS, pl) = 0, since parity conservation forbids a
one-photon Al = 0 transition, so C l, 2 vanishes.
We turn to the second order coefficient of the 2S state,

s2)=- fdt" dt'

n,p'
(2S, P2 H'(t") j nP,p' )e-i2s,2-nPan t-to)

x (nP,p' I '(t') I 1S, pi)e-iWnPPn-1sP1(t1to). (B.4)

The sum extends over all P states, including the continuum, and center of mass
momenta. We can recast this expression in the form

sp2 = f dt"V2s,P2 ;1s,p1 (t")e-iw2s,p2-is ~"t

),0
(B.5)

which is in the form of Eq. B.3. Once we have done this, we can forget about many
of the complications that arise from the two-photon nature of the transition.
To derive Eq. B.5 from Eq. B.4, we make the rotating wave approximation and the
adiabatic approximation [79] (IdEj(R)/dtj < jwEj(R)). Using the electric dipole
approximation, we set e kir to unity, but we leave eiki'A since it contains important
information on momentum exchange during the excitation. We also specialize to the
case of a standing wave laser field, for which k, = -k2= klaser, and w, = W2 = Waser,

and hlWiaser + W1S,nPI > I -

After some lines of calculations, this gives

(t) - 42 dt"(2S, P21

,92 4$2 n,pl i (wiaser + WIS-nP)

x [E 1 (R)eiklaser'R i . F1 nP, p'.) (nP,p'n F- 1 ()e-ikaser-R

n , p')(P p', i F.-2 E 2 (R)eikiaser R

+E(R)e-ikaser'R i - e
+1 -+ 2]e2iwiasert" iS, pl)e-i2spi- 1 sP2(t"-to). (B.6)

139
 To simplify the matrix elements in Eq. B.6, it is important to recall that eiki'R
is the operator for a translation in momentum space by hki. It ensures that the
atom takes up the momentum of the photon. One can expand Ei(R) in spatial
Fourier components, each of which causes momentum translation [60] as well. The
latter effect we will neglect in consideration of the matrix elements of Eq. B.6. It is
equivalently treated by viewing Ei(R) as a c-number whose amplitude varies in time
as the particle moves along the trajectory, R(t).
We can now write,

(2S, p 2 I Ei(R)e-ik.aser i - i nP,p, (nP,p, Ii - j Ej (R)eiklaserR I 1s, pi) =

Ei(R)Ej(R) r2s,nP - &i rnpJS -j 6
p 2 ,p 1 +hki+hkj. (B.7)

where rij is the dipole matrix element between states i and j and the Kronecker
delta function ensures momentum conservation. We arrive at the central result of
this section,

V2s,is(t) = h (37r 2
hc { [Ii(R)6P 2 ,P1+2hkiaser + I 2 (R)6 P2,Pi-2hkiserlM2,lS/2

+ I1 (R)I 2 (R)6P 2 ,P1M sisle2iWIasert, (B.8)

where I, = EoclEi(R) 2 is the intensity of beam i. The sums over dipole matrix
elements are reduced to

M2S, 1S =
3
2Ro 37r132
(Jo) a ((r2S,nP
so~hn (Wlaser + WIS-nP) ' i rnPJS ' @j - r2S,nP ' Fj rnPS i.

(B.9)

The term proportional to M21 ,Is gives rise to Doppler-free absorption of two counter-
propagating photons, for which there is no momentum transfer and thus no recoil
shift. The term proportional to M 1sis gives rise to recoil-shifted Doppler-sensitive
absorption of two Jo-propagating photons. The photons can come from either of the

140
two laser beams. In our particular experimental arrangement, i = 2 and w1 = W2
Wlaser = ws-is/2, and M2is M ,iS = 11.78 [80].
The AC Stark shift, AEAC Stark = h(Aw2 s - Aw 1 s), is found from an exten-

sion of this analysis. One must calculate the level shifts that arise from transi-
tions IS --+ nP -+ IS and 2S --+ nP -+ 2S [79], which we have neglected here.

For our purposes, it suffices to know that the transition frequency is shifted by
AEAc Stark/h = 3.34/ 1112Hz W- 1 cm 2 .
We now have all the pieces of the effective two level Hamiltonian, which is similar
in form to the interaction for a one-photon transition.

H' = iWS 2 (B. 10)

V2Ss hAWIS

B.2 Numerical Calculation of the Spectrum

The numerical calculation of the spectrum, Eq. 4.31, can be simplified through sym-
metry arguments and approximations. Because of the axial symmetry of the problem,
spatially, p only depends on the initial axial position, z, and distance from the z axis
(which we call r instead of p to avoid confusion with the density matrix). Also,
typically the laser pulse length is set to 400-500 /-s. In this time the atoms in our
evaporatively cooled sample move a maximum of a few millimeters. The axial length
scale for variation of the laser waist or sample density is centimeters, so we can neglect
the axial velocity.
There is another symmetry in the problem. Axial angular momentum, L, is con-
served in our trap, so we can describe the motion in r through the effective potential

Vef f (r) = 2mr 2 + U(r). (B.11)

which includes the centrifugal barrier, L 2 /2mr 2 , and the magnetic trapping potential,
U(r) =I BO + (r ,B) 2 . Here, p is the Bohr magneton, B 0 is the axial bias field
and 0,B is the linear radial field gradient produced by the quadrupole magnets. (B

141
 A

- -

r r
min max.

--- L2/2mr 2
- U(r)
'\ - Vff(r)
\ eff~r

r
Figure B-1: Energy diagram for radial motion of an atom in the magnetic trap
including the centrifugal potential, L 2 /2mr 2 . The magnetic trapping potential is
U(r) = pB2 + (r&rB) 2 . Here, p is the Bohr magneton, BO is the axial bias field
and 0,B is the linear radial field gradient produced by the quadrupole magnets. For
a given motional energy, the particle moves between the turning points at rmin and
rmax-

is really quadratic in z to provide axial confinement, but it is not important to the

discussion.) The effective potential for r motion is shown in Fig. B-1. For a given L

and total motional energy E, the particle moves between the turning points at rmax

and rmin. The evolution of the density matrix, described by Eq. 4.28, only depends on

r. For a given z, all atoms on radial orbits with a given total energy and axial orbital

momentum contribute equally provided that the laser pulse is long compared to the
radial period of oscillation for atoms in the trap. (Typically, the period for radial

oscillation is a few hundred microseconds, and the laser pulse is 400 microseconds or

more, so the approximation is not so bad.) We can parameterize the orbits by E and

L or, equivalently, the axial distance ro at which the radial velocity vanishes, and the

tangential velocity at that point, vo. (This double counts orbits because the radial

142
velocity vanishes twice in any orbit, but this is just an overall normalization.) Then

00 00 00 u (ro,z)
e
_m_0

Sv S == fJv 0 dz f wr27Tdro ro f dvo g(ro,vo) no(z)e kBT

0
2kBT
1 zo 0 rjdo ze

X p2S,2S (V, ro Z, V 0 , Tiaser), (B.12)

where we have put in the Maxwell-Boltzmann weighting of an orbit, which should

only depend on the energy of the orbit and can be related to the initial position and
velocity. We must determine g(ro, vo), the relative density of orbits at v0 and ro. If
we let P2s,2s = 1 for all orbits, then the integral over all orbits must equal the number
of atoms in the sample, which implies

J 0a 00 M U(ro,z) _ _0

S(V) = dz] 2,dro ro dvo v0 noe kBT e 2kBT

0 o kBT
X p 2 S, 2 S (v, ro, Z, vo, Taser). (B.13)

This expression gives the number of 2S atoms at the end of a laser pulse at frequency
v. With a 333 MHz Pentium II microprocessor, the code used for this thesis takes
about 24 hours to calculate 20 to 30 frequency points. This recipe must be followed
when the atom cloud is not much larger than the laser beam waist.
When the thermal radius of the sample is much greater than the laser beam radius,
the calculation can be further simplified. For all but a negligible number of atoms,
the variation of the trapping potential is insignificant in the region of the laser beam.
The trajectories of atoms which contribute to the spectrum can be described by the
impact parameter r, and the velocity when passing near the origin, v. This does not
allow for angular momentum, but atoms with significant angular momentum do not
pass through the laser and thus do not contribute to the spectrum. The contributions
from different impact parameters all have the same spectral shape as discussed when
describing the analytic derivation of the time-of-flight lineshape in Sec. 4.3.2. So the
spectrum is found from looking at the flux of atoms with a given velocity at the origin,
rather than integrating over initial conditions of orbits. This leads to an expression

143
similar to Eq. 4.24,

2
1m 2 U(r=O,z)
-00 mV
S(v) ~ 4wono dz dv v k e kBT e 2
kBT
-OC 0 kBT
X p2S,2S(V, Z, V, Taser) Tlaser , (B.14)

where P2S,2S(V, Z, V, Taser) is found for an atom initially at r = 0 with velocity v.

TOc(v) is the period of the radial motion for an atom with peak velocity v. The
factor of Tos(v)
Tlaser
is necessary because the integral counts the flux of atoms with a given
velocity, and during the numerical integration to find P2s,2s(V, Z, V, Taser), the same
atom passes through the laser beam Taser/Tosc(v) times. As written, Eq. B.14 gives
the number of excitations per second of laser exposure. This simulation program
requires an order of magnitude less computation time than Eq. B.13. The advantage
of Eq. B.14 over Eq. 4.24 is that the former can account for the effects of repeated
passes through the laser and any additional spectral broadenings or shifts, even if
they vary spatially.

144
Appendix C

Boltzmann Transport Equation

Derivation of the Cold Collision
Frequency Shift

One can explain the cold collision frequency shift from a collisional, rather than mean
field approach, using the formalism developed for cold collision frequency shifts in hy-
drogen masers [15] and more recently applied to observations in atomic fountain clocks
[16, 17, 18]. This gives additional insight into the collision events and is more easily
expanded to include inelastic processes. This section is not meant to offer a deriva-
tion from first principles; instead it applies the results of the group of B. J. Verhaar
(Eindhoven University, The Netherlands) to the 1S-2S transition in hydrogen. It is
not clear how to apply this formalism to a condensate or a inhomogeneous system,
so we restrict ourselves to the case of a homogeneous non-degenerate gas.

CA Evolution of the Single Atom Density Matrix

One starts with the equation of motion for the density matrix, p,

1 d d
p = i-[H, p]-i+ coil + rel

145
Where jp |co contains the effects of collisions and p re contains all other relax-
ation effects. (From now on we will drop the relaxation term since it is not relevant
to the discussion.) The Hamiltonian, after making the dipole and rotating wave
approximations, can be expressed as

h WO QRabi(r)e-tiasert
H = (C.1)

We have restricted ourselves to a two level system consisting of the F 1, mF 1

states of the 2S and IS levels and hwo is the level spacing. (One could easily include
other hyperfine levels which would be required to handle inelastic collisions.) For the
off diagonal coherence term we obtain

1 i d
Pls,2s = j(hwo)PiS,2S - - (PiSiS - P2S,2S)pRabi(r)eiwasert + dCt1
-1S,2S (C.2)

When jp |cn= 0, one normally makes the ansatz of PIS,2s = POs,2 seiwot. This
gives

P1S,2s - (Pis,is - P 2 s, 2 s)QRabi(r)e(wrser-wo)t. (C.3)

From this one sees that if Wiase, = wo, the fast time dependence of the differential

equation is satisfied and (pis,is - P2s,2s)QRabi drives an oscillating dipole moment

(~ Im(pIS,2s)) which, in the equation for f2s,2s, drives population into the 2S state.
In anticipation of a frequency shift and broadening arising from AP1s,2s coil, we
instead set PIS,2s =- P1s,2sei(wo+jwrt.Ti yed

dd
bIS,2s = -1w Ps,2s -2 (PISJis - P2S,2s)QRabi(r)ez~lsr-wo6+r]

+ei(wo+6+id)t PIs,2s |col (C.4)

Then to resonantly create an oscillating dipole moment, we set wiaser = wo + 6w, and
we can identify the shift and broadening through (i6w - F)PIs,2s = dP1s,2s |coll.

146
C.2 Quantum Boltzmann Transport Equation
The collision term can be found from the Bogoliubov-Born-Green-Kirkwood-Yvon
hierarchy[132] for the quantum Boltzmann transport equation[15].

d
jPIS,2S |coll (i6wO - IF)P1s,2s =

6 (C.5)
P1s,2s E Z[(1 + 6Sp)(1 + 62s,p)(1 + is,v)(1 + 62S,v)]1/2(VO(1S,2S),(v-+[,)).
V /

Brackets refer to a thermal average over velocity, v = ( ). k is the momentum of

each particle of mass m in the center of mass frame. This is a local equation, so nu
is the local density of atoms in internal atomic state v. It inherently assumes that
the relative populations in the states don't depend on atomic velocity. 0(1S,2),(v-+P)

is the "cross section" for an atom in a superposition of IS and 2S to collide with an

atom which makes a transition from v -+ p. This is given by

U(1S,2S),(v- p) = Z(l + 1)[Stls,L],[S,v] S[2Sp],[2S,v] - Pv].

s ,s is the i-wave scattering matrix for the collision between the properly sym-
metrized wave functions for the incoming state of two atoms, [IS, p], and the outgoing
state, [1s, v].

C.3 Application to the 1S-2S Transition in Trapped

Hydrogen
Now we can specialize to our experimental situation. We restrict ourselves to s-wave
collisions, and the only nonvanishing density is that of the IS atoms, so v is set to
IS. Inelastic collisions contribute through cross sections such as

U(1s,2S),(1S-+) [Ss,p),[ISIS]-[2Sy),[2SIS] - 1

147
where p ranges over the IS hyperfine levels, a, b, and c, but no relevant inelastic
collision cross sections have been calculated. The collision term arises from

0-(1s,2S),(1s*1S) = [Sis,is],[is,ls]S[2s,is],[ 2 s,is - 1]. (C.6)

This yields

d PIS,2s 1col= (i6w - F)PIs,2s = PIs,2snIs 2 K mk) 0(1 S,2S),(ISIS) (C.7)

In the zero temperature limit, the s-wave elastic scattering matrices are given
by Sos is],[is] exp[-2ikais,is] and So0sis][ 2 sis= exp[-2ika2 s,isl. This implies
that
d
PIS,2S lcoI= (i6W - F)PIs,2s =

PIs,2snis2 ((2hk) (7k) [(1 - 2ikais,is)(1 + 2ika 2 S,Is - 4k 2 as is) - 1])

= P1s,2S I- (
87rhnls
(a 2 s,is - ais,is) - nis87ra2sis
2S
K
(2
k) (Ck\
(C.8)

C.4 Discussion

We can identify the frequency shift in the first term in Eq. C.8 as 6w = (8wrhnis/m) (a s,IS-
2

ais,1 s), which is identical to the expression found using the mean field approach

for a homogeneous gas with homogeneous excitation probability. The second term

is the next higher order term in ka, which contributes to the broadening, F =

nis87rasis(reative). The broadening is equal to the normal collision rate for atoms

in the 2S state with IS atoms assuming o- =87a 2 as is the case for collisions
between identical particles. iS and 2S particles are distinguishable, but somewhere
buried in the quantum Boltzmann equation is the same 1S-2S exchange effect de-

scribed in Sec. 6.1.3. Evidently, the formalism assumes all motional states are equally
excited, so it may need to be modified to describe an inhomogeneously excited sample.

We can gain some intuition from these expressions. If the phase shift per collision

148
is large, then the line is mostly broadened. The phase of the oscillating dipole is
randomized in every collision, so the atom cannot determine the frequency of the
radiation to better than IF. If the phase shift per collision is small, however, the
dipole never completely loses memory of its phase relation with the electric field. The
net effect of the collisional phase shifts is to decrease or increase the natural frequency
of revolution of the dipole, depending upon the sign of the scattering length, which
changes the resonant frequency. The shift in this case is larger than the broadening.
The thermal average of kals-2s is 27rais-2s/AT, where the thermal deBroglie wave-
length is AT = h/V/27rmkBT. For a typical trapped hydrogen sample, T ~~ 100 PK,
AT = 123 nm and the broadening is around 0.18 of the shift. The importance of
the broadening decreases with decreasing temperature. It is small compared to other
linewidths in the spectrum, but not entirely negligible.
It is important to note that the equations for the time evolution of the single parti-
cle density matrix, which form the basis of the numerical simulations described in Sec.
4.4, are the same whether one includes a mean field energy shift in the Hamiltonian
or the collisional term jps,2s coI-

149
Appendix D

Details of the Mean Field Theory

Calculation of the Cold Collision
Frequency Shift

This appendix provides detailed calculations of some of the results stated in the cold
collision frequency shift theory chapter. For ease of calculation, more explicit notation
is often used.

D.1 Correlation Functions for a Homogeneous Sys-

tem

The notation is described in Sec. 6.1. The normalized second order spatial correlation
function of the gas, g( 2 )(x), as defined in Eq. 6.11, is

g( 2 )(X)
I dr {
itA
(TN;1(p - r) -
-(i r - x VN;O / [n (r)]2 (D.1)

where n(r) is the density at position r,

N
n (r) - Z XFN;O16(i r) I4 JfN;O) (D. 2)

150
and the state vector for N IS particles, 41 N;o), for a noncondensed and condensed
gas, is given in Eq. 6.2 and 6.3.
For the case of particles in a box of volume V, the motional states can be taken
as plane waves with periodic boundary conditions, (r~k) = exp(-ik - r)/V'V, so

(kik2 |6(i 1 - i 2) k3k4 ) = 6ki+k2,k3+k4, (D.3)

and n(r) = N/V. Equation D.1 then reduces to

2
()=v2 N
g (X) N_ Z(XpN;O 16(i _ j + X) I4 1fN;o). (D.4)
N2i:j

The state vectors are symmetric with respect to particle label, so

(2) g~x) _ V(NN- 1) (I;~(N;0 2 ++x~')

X) IN;O). (D.5)
2

We now specialize to a non-condensed gas with no state doubly occupied. For

x = 0, the correlation function reduces to

N
V(N -1)
(0)
NN! E (kQ(l)kQ( 2)|6(ri - i2) kR(1)kR(2)) 11 6 kQ(j),kR(i)
Q,R i=3
(D.6)

For all i > 3, for every nonzero contribution to g( 2 )(0), kQ(i) = kR(i) due to the Kro-
necker delta functions. Thus, for each permutation Q, 2 R's contribute - a direct term,
(kQ(j)kQ( 2 )|6(iI -i 2 ) kQ(l)kQ( 2 )), and a momentum exchange term, (kQ(i)kQ( 2 )16(f1 -
F2 ) kQ( 2 )kQ(l)). Thus

V(N-1) ((kQ(1)kQ(2)|6(i1
gorm(0) - i 2 ) kQ(l)kQ( 2 ))

+(kQ()kQ( 2) 6(fi - i 2 ) kQ( 2 )kQ(l)))

V(N-1) 2 2(N-1)
(D.7)
NN! V N

151
 For a condensate

(2)
YBEC (0)
V(N - 1) (k = 0 k2 = 016(i - r2 ) k= 0 k2 =0) f Ok0NO
Ni=3
V(N-1) (N-1)
(D.8)
NV N

D.2 Interaction Energy for a Homogeneous Sys-

tem before Excitation
The interaction Hamiltonian, H', is defined in Eq. 6.8. The interaction energy, E/N;o
for a condensed or noncondensed homogeneous gas, as stated in Eq. 6.10, is given by

EIN;O _ (N;O H'lN;O)

N 471h2 ais-is
m 6(i - N;
(4 N;0
1<i<jm
27Fh 2a s-3 N
1
(4
-iS N;O Z(ii _ ij) IqN;O)
II
i/zj
2
27h ais-is N 2 (2)
(0). (D.9)
m

D.3 Interaction Energy for a Homogeneous Sys-

tem after Excitation

For a homogeneous system, the state vectors after excitation for a noncondensed and
condensed gas are given in Eq. 6.5. The calculation of the interaction energy is simpler
if they are expressed, for both a condensed and noncondensed gas, as

n!(N - n
N ) (2S)(1iS ) 1(|2S)(1S |) 2...(2S)(IS ) II N;O). (D.10)
{N }

The operator (12S)(1IS)i changes atom i from a 1S to a 2S atom while leaving its
motional state unchanged, and {u-} refers to an allowed set of o-i's, (1 < O-1 < o 2 ... <

152
Un < N). The sum over all {} reflects the fact that we are equally likely to excite
any atom, and it produces a state which is totally symmetric with respect to particle
label and the motional states of the 2S atoms.
Using Eq. D.10 and 6.8, we can write the interaction energy as

EIN-hf, _
-fNiIHIIN-A,Ai _h!(N - f)!

x (WN; N 2-!H'

E(j2S)(IS )r |2S) (1SD)... 12S)(1SI) jqN;O

.I /} !
= !(N - )!
E{} N;0 (2 -- O
N!

EWl,mn (12S)(1S r,I 2S)(1S3-). 12S)(1SI )ii II N; 0

1<1<m
h!(N - h)!
N!
N n n
(25Io~ IN;O)
X E E ({},N;{ l"mI H(2S(S
15<M fu},{,} L 1 -1 .
(D.11)

Because wI,m is diagonal in the internal states of atoms 1 and m, only terms with
{u} = {} are nonzero. The sum over {a} and {} reduces to a sum over the

(N) (N! -different {}'s.

EIN-h,h _ i!(N - A)!

N!
Nm
x E E (qN;O WI,m L iN;
15<m {u}
(D.12)

For a given 1 and m, there are four classes into which the various {} fall. There
are (N-2) N - permutations for which no u- takes the value 1 or m and the
interaction is iS-iS. There are (N-2) !NNpermutations for which

153
one o-i takes on the value 1, and none are m, and there are (-2 permutations for
which one -i takes the value m, and none are 1. For both these cases the interaction
is 1S-2S. Finally, there are N-2 2 (N-2)! permutations for which one -i
(n-z 2 (t22)--fi2)

takes the value 1, and a o-j is m, and the interaction is 2S-2S and vanishes. This
implies

E N-ii _ (N - h)!
- N -0
N! 1<1Km (N-- 2 2)!
h! (N - fz)! N

(N - 2)!
- 1)!(N - 1 - I)
(AT - 9'
+ _
(n-1)!(N - I - h)!
- (WN;0 1(IS)K2s )2 W1,ml (12S)(1S)2 4N;o)
I
(D.13)

We can insert the form of wl,m from Eq. 6.8 and evaluate the matrix elements of
the internal states to find

EiN-hh
fl! (N - fi)!

N
N! F (N - 2)!
n!N-

47mh 2
2 - n)!ais-1s +± (n
2
-
(N -2)!1
1)!(N - 1 - n)!ais-2s

x E (FN;-O q/N;O)
1<1<m
27h 2 (N - ) [(N
- - l)ais-is + 2nas-2s]
m N(N - 1)
N
X (4 pN;O 16(i im) IpN;O)
l~m .

27h 2
N(N - h)
- - 1)ais-is + 2naIs-2s1 g(2 )(0). (D.14)
m V(N - 1)[(N

This result was stated in Eq. 6.13.

154
D.4 Derivation of the Energy Functional for the
Excited State of a Condensed Gas in a Trap
In this section we derive the energy functional for the excited state of a condensed
gas in a trap, Eq. 6.26. See Sec. 6.2.1 for a full description of the notation.
The energy of the excited state after laser excitation, and the 2S motional wave
functions, are found by minimizing the energy functional,

-Gp'= (IN-P;P'i H1 N -Pi), (D.15)

where

N-p,2S; N-p,2S. IS, (D.16)

WN-p;p,i) S ___2S, _____... __2S__ 0.N-p,1S . i N-p,1S),
p terms N-p terms

and k> p,2S is the 2S motional state which is resonantly excited. The symmetry
operator is S = pQ,-p)! 'y Q where the sum runs over the N'! particle label

permutations which produce unique kets.

The Hamiltonian is defined in Eq. 6.18 and the energy functional is

EB EC'" = qN pp,il HJ jpWN-p p,i)

= N-p,i j + U(ij) + Hjn ) + H' lq -Gp'i)

p2
= (N -p)(SON-1S U(i) + Hint |1S,0oN-P,1S)
2m +

+p( 2 S, k -,P, 2 I (2m + U(i) + Hint) 2S, k -P,2S)

+(BE Gp' H' KN -PPi). (D.17)

We will break this into pieces and evaluate the interaction part,

( EC 'H' - ' ) = (...|SH'S|...

155
 = (...H'SSI...) (...JH'S j...), (D.18)

where we have used the fact that H' is diagonal in the space of the internal states of
the atoms, and SS S. Inserting H', we find,

47rh 2 [
(... E 6(ij-ijf){as-lspiSSp1S + a1l-2s P/S + P2SP1S] S1...). (D.19)
M1<z<i

Of the permutations in S, (N-2p of them result in a IS - IS interaction 2(N-2)!

of them result in a IS - 2S interaction, and the rest result in a 2S - 2S interaction
which vanishes. The expectation value of H' reduces to

2
p!(N - p)! N(N - 1) 47h
N! 2 m
(N -- 2 )! als-1s(ON-p,1; ON-p,1S1
- i 2 ) ON-p1S; ON-p,1S)
p!(N - p - 2)!
+ 2(N - 2)! -S~fp2S;ON-p,1S
N 1 2 p,2S; ON-p,1S)
(p - 1)!(N - p - l)
27r-h2
= (N - p)(N - p - 1)as-is(ON-plS; N-p,1S 16j - 2 )ON-p,1S; 0 N-p,S)

4wrh 2
+ p(N - p)a1S-2S(k-p,2S; 0 N-p,1S 6(i 1 - i 2) kp, 2
S; 0 N-p,1S). (D.20)
m

Inserting this result into Eq. D.17, we find the energy functional is

EN--p;p,i
BEC
=E N-p;0
BEC

+p(2S, k{-P, 2 S Hi"l + + U(i) + 47rh2 aS-2S N-p 2 -P,2S)

2m m
= EB 0 + p(EIS-2S + Ei), (D.21)

where E N- 0, as defined in Eq. 6.23, is

ENEp0= (N- p)(1S,0N-p,1S| ($2 + U(i) + Hint) |1S, 0 N-p,1S)

2
2 _rh
+ (N - p)(N - p -1)als-1 s
m
x ( 0 N--PS; 0 N-p,1S3( 1 -~ 2 N-P,1S ON-P,1S), (D.22)

156
and

n C-(r) = ( 0 N-p,1S 1 6(i - i 0 N-p,1S) (D.23)

is the density in the condensate for N - p condensate atoms.

D.5 Details of Elements of the Proof of the Sum

Rule for the Mean Frequency of the Spectrum
for an Arbitrary System

In Sec. 6.3 we proved a sum rule (Eq. 6.33) for the cold collision frequency shift of
the spectrum for an arbitrary system. Here we provide more detailed calculations of
some of the intermediate results. Refer to Sec. 6.3 for a complete explanation of the
notation.
For the collisional interaction, we use the Hamiltonian, H', given in Eq. 6.8, except
we set ais-is = 0. H' is written

H' = 47rh2 ajs2s N

(D.24)
M SmAn
1 6(ifm - in)P 2 SP1S-

For the state of the system before excitation, we take a general state with N
IS atoms, denoted by ITN,o). This state could describe bosons, fermions, or even
classical particles [110], in any thermodynamic state.
For the state after laser excitation we consider all configurations of the system
with 1 2S atom and N - 1 iS atoms. In the absence of H', this manifold of states is
degenerate, with energy about equal to EIs- 2s (neglecting kinetic energy). H' lifts
some of the degeneracy, and one can then think of the laser exciting the system to
one or a distribution of the eigenstates of H', as shown in Fig. 6-8. We denote the
eigenstates as jvi), and H'Ivi) = Eflvj).

157
 We can write an expression for the spectrum,

S(2hv) = | (Vi|HiasIN;0)26(2hv- EIs-2s -E|), (D.25)

where the sum runs over all eignestates Ivi). The overlap matrix element is found
from the atom-laser Hamiltonian which can be written (Eq. 4.8)

N hQO 0 j) (I2S)(IS ),
Hias E 2 (D.26)
j=1

where Qo(r) is defined in Eq. 4.17 and the sum runs over the N particles.
We can show that the spectrum given by Eq. D.25 obeys the sum rule, Eq. 6.33.
We start with

J 2hdv(2hv - E 1 s- 2s)S(2hv)

2hdv(2hv - E 1 s- 2s) 2 I(Vi|Hias|

1N;O) 26(2h - Els-2s- Ej)

-z E (N;|HtasIvi)(vi Hias |, N;0)

2- N;O iasH'|vi)(vi Ias|IJN;O).

The eigenstates v2 ) are a complete orthonormal basis for the Hilbert space connected
to 1,FN;o) by Hias, so the expression further simplifies to

2w (4N;OHtsH'HIas
l IN;0
h
hQ 0 (Rj)(IIS)(2SD
= ( N;01 1
j=1

x 47rh 2 aS-2S . 6(m

hQ2
(fi 2S)(1S ).
- n )p2Spis
m m~n
m=1 2 (2S IS1 N;O)

27r 47h 2 aIS-2 ( , N;O hQ 0(m)

(iS) (2S);
h MiS2 2

X6o(m- in)p 2 SpiS hQO(Rm) (12S)

2

158
 21 47rh 2 as-2S N( N;O hQO M)6 (im - hQ 0 I N0
h m m: 2 2
27r 47rh 2 ais-2s
d3ri d 3r 2 ri r1m
h m f E(n m
m:An2

X6(irm - 2 m r, JilqN;0

2 47rh 2 a 1 s-2s 3r 1hQo(ri)- 2 N;O r)(rm(lr)(ri a N;

h m I . m~kn jj r jm jr)r jn'

27r 47h 2 as-2sfd

F3 r Eo(ri) 2
(N;0 6 (m - r)(- ri) N; 0 )
h m I
47h 2 ajs-2s 2
[d3r -rhD0(ri) [n(r) ]2 G(r, 0).
(D.27)
m I 2

In the last line we have identified the correlation function, G(r, 0), defined in Eq.
6.34.
We also need to calculate

J 2hdvS(2hv) =

|I:hI (VJ|HIlas| N 126 (2hu - Ejs- E )

= f 2hdv
N 2
;|
- h (~~Has IVi) (Vi JHias IqN;o)
2w
(,N;O HtasHias JIN;O)
h
N N
27r 2
I 2N; N;0
j=1 =
2w N hQ0 (ij) hQ 0 (ij)
- 2((1N;S)(2SI)i 2 (12S)( 1Sj)iN;O)
i=
2w N
N;Z
-2 ;
N
2w
-w Jd3
2 1
(jr(rjjj N;O)

2w
hQo (r) 2
-w d3 r S(4pN;0 (Ir) (rI) 4N;O)

2- £Zd 3 EhQo>Z'(i 2
(r)1 (qIN;O1
21]
N
- r) IXJJN;O)

159
 J d3r 2 n(r). (D.28)

Taking the ratio of Eq. D.27 and D.28 and recalling that QO(R) oc I(R) (Eq. 4.17)
proves the sum rule, Eq. 6.33, within the approximation that ais-is = 0. It is
straightforward to include ais-is in this proof.

160
Appendix E

130 Te 2 Reference Spectroscopy

This appendix describes the 13'Te 2 reference spectrometer used to locate the frequency
of the two-photon 1S-2S F = 1 transition within a few hundred kHz. The 130 Te
2
cells were obtained from Opthos Instruments, Rockville, MD, (301)926-0589.

E.1 Introduction
Detecting the two-photon 1S-2S F = 1 signal in this experiment remains a challenging
endeavor even though it has developed into a working tool. The search, especially the
first time attempted during a cool down, is in a phase space with many parameters.
Laser frequency and spatial overlap of the laser and the atom cloud must be right,
and the many components of the experiment must all function (and be turned on).
We typically search for the signal in a low compression trap, which makes laser-
atom overlap easier to obtain. We also use a sample with a temperature of around
20 mK so that the sample preparation time is minimized due to the short evaporation
route. The 1S-2S Doppler-free line is still only 20-30 kHz wide at 243 nm, even in
this warm trap. The best commercial wavemeters can only fix the frequency to a few
hundred MHz, which leaves a large search range. The solution to this problem is to
utilize the well-studied absorption spectrum of 130 Te as a reference. The frequencies
2

of many lines in the 486 nm region have been determined to an accuracy of about
1 MHz[133, 134] because of the interest in 1S-2S spectroscopy of hydrogen, deuterium,

161
 I 1i 1 2.0
1.5 -

12841.6
, 1.2 -
LXJ

- 1.2
S0.9 - b
CC

0
- -0.8.
0.6

0.

'D0.
Cn =3

0 0

-15 -11 -7 -3 1 5
Frequency Offset [GHz]
Figure E-1: Doppler-sensitive spectrum of 130Te2 - upper trace. Saturated absorption
spectrum of same region - lower trace. The Doppler-broadened line 1284 is often used
to calibrate sample temperature. Lines b2 and i 2 are important references for 1S-2S
hydrogen spectroscopy.

muonium and positronium. Relative measurement of the difference frequency between

a 130Te line and a particular 1S-2S line can be substantially better[135].
2

Spectroscopy of 1 30Te is relatively easy. The substance is a gray metal at room

2

temperature, but has a substantial vapor pressure at around 500 'C.

The simplest method to interrogate the spectrum is with direct absorption of a
probe beam[136]. This yields Doppler-broadened lines of the form

S(,Av) ~ Ce kT v, (E. 1)

where vo is the transition frequency. The resulting linewidth, FWHM, is about

760 MHz at 486 nm. It is very difficult to derive line centers from such features

162
with 1 MHz accuracy, especially when there are many spectral lines contributing
to a single Doppler feature, as is often the case. Figure E-1 shows an example of
Doppler sensitive spectroscopy of Tellurium. One can avoid the limitation of the
Doppler width by performing saturated absorption spectroscopy, as is also shown in
the figure.

E.2 Saturated Absorption Spectroscopy

The full theory of the sub-Doppler technique, saturated absorption spectroscopy, is
complicated[137], but a simple model will suffice here. A pump beam along the z-
axis of the sample burns a hole in the z velocity distribution of the ground state
by exciting atoms which are at the velocity which is brought into resonance by the
Doppler shift. (Complete saturation of the transition reduces the population to 1/2 its
value.) The width of this feature in the velocity distribution is nominally 7natural/k
27F is the natural linewidth of the transition in Hz (typically 10-20
where 'naturai/

MHz) and k is the wavevector of the laser, although the feature can be broadened
a bit by power broadening. For a few mW of power, this feature is much narrower
than width of the entire 500 K distribution. The Doppler sensitive absorption of a
counter-propagating probe beam essentially maps out the z velocity distribution of
the sample. It shows the normal Doppler profile, except for a decrease in absorption
corresponding to the velocity class excited by the pump beam. This dip in absorption
is the saturated absorption signal.
Normally, both pump and probe are derived from the same laser, so the frequencies
of both are scanned simultaneously. In this configuration, the saturated absorption
feature has a width equal to half the natural linewidth.
The presence of the dip on the probe absorption indicates that both beams are
interacting with the same velocity class. Atoms in this class have z velocity within
S'Ynaturai/k of v2, where vz must satisfy vo = Vpump-kvz/27F and vo = Vprobe+kvz/27r.

This implies Vprobe + Vpump= 2vo. This shows how the Doppler effect cancels and also
that the signal arises from only a small fraction of the atoms in the laser beam,

163
 control con

Ring Dye Laser cavR

EOM
Kr+ Laser .
EOM
Coherent 699 Teinp.
inrnova 200 486 nm control

AOM To Doubling Cavity

LA~K)

1 3Te
2 I 50/50

Probe Oven Pump

laser T llaser 6
Te Power Monitor
Ternp.
control

Figure E-2: Components of the laser system which are important for Tellurium spec-
troscopy. Beam splitters are 4% surface reflections unless otherwise noted.

' Ynatural/YDoppler, as opposed to the signal from Doppler-free 1S-2S spectroscopy,

which arises from all the atoms. In practice, the pump beam is amplitude modulated,
which modulates the saturated absorption feature. One can then use lock-in detection
to pick the small saturated absorption signal out of the large Doppler background.

E.2.1 Laser System

The optical layout is shown in Fig. E-2. The frequency of the dye laser can be
controlled by its commercial optical cavity, producing a 1 MHz linewidth and allowing
scans of up to 30 GHz. Alternatively, it can be locked to a mode of the stable reference
cavity. The linewidth is then less than 1 kHz. The frequency of the cavity modes are
fixed, so the frequency of the laser in this configuration is scanned by scanning 6 cav,
the frequency of the AOM which splits off the beam that goes to the reference cavity.

164
 3.5 8

3.0 7

X1/4 1S-2S F=1-

2.0 -

2 1.5 -
2 c
1.0 3

0.5 -

0 0

-0.5
-1000 -800 -600 -400 -200 0 200
frequency [MHz]

Figure E-3: Saturated absorption spectrum near 1/4 of the hydrogen 1S-2S F 1
transition. The lower trace is the transmission of the reference cavity which provides
the frequency calibration.

Because the first-order AOM deflected beam must remain aligned with the reference

cavity optics, the tuning range when locked to the cavity is limited to 20 MHz.
The pump beam is frequency shifted by 6Te from the laser frequency, Vlaser. This

makes it easy to modulate the pump beam at 20 kHz and allows offsetting of center

frequency of the saturated absorption signal. As the laser frequency is scanned, the
6
center of the saturated absorption signal is at Viaser = vo - Te/2. The pump beam

has a waist radius of 500 pm in the tellurium cell, and a peak power of 30 mW. The
probe beam has a waist of 400 pm and a power of 5 mW. After passing through the
cell, the probe beam falls on a fast photodiode. The saturated absorption signal is

extracted by a Stanford Research Systems lock-in which demodulates the photodiode

signal at 20 kHz with a time constant of 300 ms or shorter.

E.2.2 1S-2S F = 1 Reference Transition, i 2

The most useful " 0 Te2 line for 1S-2S spectroscopy of hydrogen is labeled i 2 in Fig.

E-3. This spectrum was recorded by locking the laser to the commercial cavity and

scanning with the Coherent 699 internal scan. The location of the hydrogen line is

165
 1
19gMay1 533.te2 - 3Te 2 i, Saturated Absorption
I I I i I I
S13 3.5
0
x2 = 0.7929
gammal = 5.589t0.024
gaini = 0.00011791t3e-07
~11 center freq = 201.416*0.0048 - 3.3
offset = 6.98e-06t3.5e-07
0
3.1 CD
at

X 9 2.9
0 75
2.7

2.5
C 3 i i
180 190 200 210 220 230
Reference Cavity AOM Frequency [MHz]

Figure E-4: High resolution saturated absorption spectrum of the i 2 line in 130 Te2 .
For this spectrum, the laser frequency was locked to the external reference cavity.
The signal is normalized by the square of the pump power, and parameters of the
Lorentzian fit are given.

indicated. Figure E-4 is a high resolution scan of i 2 obtained by locking the laser
frequency to the reference cavity and scanning with the reference cavity AOM. The
latter procedure is our standard way of recording the spectrum. We fit the line to a
Lorentzian and can split a given line to about 10 kHz. The quoted width in the fit is
HWHM. Unfortunately, the maximum scan range is only about 2 full linewidths.
The absolute frequency of i 2 is known to better than 1 MHz[133, 134]. More useful
than the absolute frequency, however, is the relative separation between i 2 and the
Doppler-free hydrogen line. This can be found with RF accuracy, and in practice we
tune 6Te so that we are on resonance with the Tellurium line and 1/4 of the hydrogen
line frequency at the same value of 6cav. Mclntyre[135] found the frequency of i 2 to lie
57.1(4) MHz above 1/4 of the F = 1, 1S-2S transition in hydrogen. Unfortunately,
the frequency of the line has a large temperature dependence, and impurities in the
cell can produce cell to cell variations of as much as 1 MHz. These systematics must
be controlled if the i 2 line is to be useful as a sub-MHz frequency standard.

166
 sj57.3
C,,
CnJ

LL 57.1

056.9

1 6.7

C.56.5 -
LIE 480 485 490 495 500 505 510
0
Oven Temperature [ C]

130
Figure E-5: Temperature dependence of the frequency of the i 2 line in Te2 at
saturated vapor pressure. The frequency is referenced to 1/4 of the F = 1 1S-2S
Doppler-free hydrogen frequency.

E.3 Systematics of i 2 Frequency Stability

E.3.1 Temperature

McIntyre found that the i2 frequency varied by -30 kHz/ 0 C, which corresponds to

-1.10 MHz/torr of saturated vapor pressure. The standard condition for which the

frequency is quoted is 513(5) 'C or 0.89(11) T. The cell temperature is difficult to

measure and it may not even be uniform. In our case, we have an uncalibrated

thermocouple gauge in the oven, a few centimeters from the cell. The temperature of

the thermocouple is controlled by a commercial servo-system, but this probably only

controls the cell temperature to about ±5 'C. Once we were able to detect the F = 1

1S-2S Doppler-free signal, we carried out several studies which should make the i 2
line a more reliable reference in the future.

Figure E-5 shows the temperature dependence of the separation of the i 2 and IS-

2S frequencies. We derive -24 kHz/ 0 C for our laboratory definition of temperature,

but more importantly, it is clear that we can control the cell temperature to the

quoted ±5 'C.

167
 If the thermocouple moves, or if the tellurium cell or the oven configuration must
be changed, it is not clear that the thermocouple temperature can reliably set the
conditions to determine the frequency within 1 MHz. Fortunately there is a more
absolute means of calibrating the temperature.

E.3.2 Column Density

The absorption on the center of a Doppler-sensitive peak is a measure of the inte-

grated column density, which is directly related to temperature through the vapor
pressure curve. The absorption of line 1284, shown in Fig. E-1, has become a stan-
dard reference. McIntyre worked with a 23(2)% absorption in a 7.5 cm cell. We have
a 10 cm cell, so the corresponding absorption is 29%. Figure E-6 shows how the
absorption varies with oven temperature. The thermal time constant for the oven is
on the order of 10 hours, so patience is required to record data such as this. The error
in a single absorption measurement is less than 1%, while the scatter at 500 0 C, and
the overall smoothness of the plot confirms that cell temperature is stable to ±5 0 C.
The temperature stability is also shown in Fig. E-7, which plots all the frequency
measurements made at an oven temperature of 490 'C. The standard deviation is
around 100 kHz.

E.3.3 Reliability and Cell to Cell Variation

While the absorption at 513 'C is consistent with what McIntyre quotes, the frequency
with respect to 1S-2S differs by about 700 kHz. Our typical working temperature
is 490 'C, which gives an average frequency separation of about 57.1 MHz and an
absorption on line 1284 of 17%.
In practice, it is good to set the temperature so that the absorption on 1284 is
about 17% and then check the linewidth of the i 2 line. The linewidth has proven to
be a sensitive probe of the purity of the cell.
Figure E-8 shows how the linewidth depends on oven temperature. There is
significant scatter in the data, implying some parameter was not well controlled (such

168
 0.35 I I I I I I

0.30 [
0 i

0.25
0

0
W2
0.20 I

I I I I I I
0.15
490 495 500 505 510 515
Oven Temperature [*C]

Figure E-6: Absorption on the center of Doppler-broadened line 1284.

I I I

"J57.25
C4,

57.20 I.
06
57.15 - U U

61

C&
57.10 -

57. 05 U U

ii

57..00

I . . I . . I
U56..95-
5.5 5.8 6.1 6.4
i2 linewidth [MHz HWHM]

Figure E-7: Frequency of i 2 plotted against linewidth. The oven temperature was
controlled at 490 0 C and the measurements were taken over a 4 month period.

169
 7.2 i I I I I I I I

6.6

a U

6.0

5.4 - -
480 485 490 495 500 505 510 515
Oven Temperature [*C]

Figure E-8: Linewidth of i 2 versus oven temperature.

as the power broadening), but there is an obvious temperature dependence. For 17%

absorption on 1284 and the light powers given above, one should get a linewidth,

HWHM of about 5.9-6.0 MHz. If the linewidth deviates from this value, especially if

it is too large, the cell should be used with caution because it might have a relatively

high level of contaminants.

For a previous data set with two other tellurium cells, we observed a tight correla-

tion between linewidth and frequency when the oven temperature was set to 490 'C.

(See Fig. E-9.) Perhaps parameters were more tightly controlled in this study. One

can clearly note that one of the cells consistently had a linewidth which was too large

and the frequency is low by about 500 kHz. Considering this data, and the fact that

we once received a "13 0 Te2 cell" which actually contained zinc, it is clear that one

needs to carefully assess the quality of uncalibrated cells.

E.4 Details of MIT Experimental Procedure

There are a few important procedures to follow in order to use the Tellurium reliably.

Always align the beams carefully. It should be possible to obtain a signal by over-

170
 I I I I
57.2

C4,

(n57.0

F 56.8

J6.6

56.4
I I I I

5.0 5.5 6.0 6.5 7.0 7.5

i2 Unewidth [MHz HWHM]

Figure E-9: Frequency of i 2 observed in two other cells, plotted against linewidth.
The oven temperature was controlled at 490 'C. Cell 2 appears to have excessive
contamination.

lapping the beams by eye, but then peak up the signal by viewing the output of the

lock-in on a scope while adjusting the mirrors.

The power levels were given above. McIntyre found that the frequency was in-

sensitive to power, but we have plenty of signal and in general it is good to work at

lower powers to reduce any possible power broadening. It might be useful to study
how the linewidth varies with power to check if power broadening is contributing to

the lack of correlation in Fig. E-7.

As the frequency is swept, the power to the Tellurium spectrometer varies due to

the need to keep the power on the reference cavity constant. The power is monitored

as shown in Fig. E-2 and Fig E-4. The saturated absorption is a two photon process,
so the lockin signal must be normalized by the square of the power to account for the
power fluctuations.

The time constant on the lockin should be 300 ms or shorter. A typical scan
consists of 300 points across 20 MHz and is recorded in about 5 minutes by a computer

and GPIB controller. A longer time constant or faster scan will distort the lineshape

enough to move the line center appreciably. A single GPIB measurement requires

171
about 30 ms. To take advantage of all that potential averaging, the program can
record multiple measurements per point. 30 is a reasonable number so that the laser
sits at one frequency for about a second. The statistics of the 30 measurements gives
one an idea of the noise in the spectrum.
One should worry about the baseline, especially since we do not scan many
linewidths and we try to get linewidth information. It should be less than about
5-7% of the amplitude as in Fig. E-4. Check that the baseline goes to zero when
either beam is blocked. Note that the Lorentzian fitting function for i2 includes the
effects of the broad line to lower frequency shown in Fig. E-3. This doesn't effect the
frequency of the i2 line that much, but is more important if one is trying to glean
information from the the linewidth and baseline. Note that if one changes the or-
der of the AOM beam which goes to the reference cavity, the sweep direction of the
spectrum changes as well. The broad line changes from the low frequency side to the
high frequency side of i 2 . (All the data presented in this thesis is consistent with the
broad line being to the low frequency side of i 2.)
In the future, studies of the power dependence of the linewidth and design of a
system to allow wider frequency scans while locked to the high finesse cavity would
be useful. This could be accomplished for example, by electronically controlling the
laser beam alignment into the reference cavity optics. In its present form, however,
the Tellurium spectrometer is robust and reliable. Think carefully before changing
the oven, thermocouple, or cell. I do not recommend changing the setup unless the
hydrogen signal is readily accessible for recalibration.
Good luck!

172
Bibliography

[1] N. Bohr. The Theory of Spectra and Atomic Constitution.Cambridge University

Press, 1922.

[2] M. Born. Atomic Physics, chapter 5. Dover Publications, Inc., 1989.

[3] G. Gamow. Thirty Years that Shook Physics, chapter 6. Dover Publications,
1985.

[4] J. E. Nafe, E.B. Nelson, and I. I. Rabi. Phys. Rev., 71:914, 1947.

[5] J. E. Nafe, E.B. Nelson, and I. I. Rabi. Phys. Rev., 73:718, 1948.

[6] W. E. Lamb and R. C. Retherford. Fine Structure of the Hydrogen Atom. Part
I. Phys. Rev., 79:549, 1950.

[7] W. E. Lamb and R. C. Retherford. Fine Structure of the Hydrogen Atom. Part
II. Phys. Rev., 81:222, 1951.

[8] K Pachucki, D. Leibfried, M. Weitz, A. Huber, W. K6nig, and T. W. Hdnsch.

Theory of the Energy Levels and Precise Two-Photon Spectroscopy of Atomic
Hydrogen and Deuterium. J. Phys. B: At. Mol. Opt. Phys., 29:177, 1996.

[9] Th. Udem, A. Huber, B. Gross, J. Reichert, M. Prevedelli, M. Weitz, and

T. W. Hnsch. Phase-Coherent Measurement of the Hydrogen 1S-2S Transition
Frequency with an Optical Frequency Divider Chain. Phys. Rev. Lett., 79:2646,
1997.

173
[10] C. L. Cesar, D. G. Fried, T. C. Killian, A. D. Polcyn, J. C. Sandberg, I. A.
Yu, T. J. Greytak, D. Kleppner, and J. M Doyle. Two-Photon Spectroscopy of
Trapped Atomic Hydrogen. Phys. Rev. Lett., 77:255, 1996.

[11] T. C. Killian, D. G. Fried, L. Willmann, D. Landhuis, S. Moss, T. J. Greytak,

and D. Kleppner. Cold Collision Frequency Shift of the 1S-2S Transition in
Hydrogen. Phys. Rev. Lett., 81:3807, 1998.

[12] B. J. Verhaar. Cold Collision Phenomena. In D. J. Wineland, C. E. Wieman,

and J. Smith, editors, Atomic Physics 14: Fourteenth InternationalConference
on Atomic Physics, page 351, New York, 1995. American Institute of Physics.

[13] D. G. Fried, T. C. Killian, L. Willmann, D. Landhuis, S. Moss, D. Kleppner,

and T. J. Greytak. Bose-Einstein Condensation of Atomic Hydrogen. Phys.
Rev. Lett., 81:3811, 1998.

[14] C. H. Townes and A. L. Schawlow. Microwave Spectroscopy, chapter 13. Dover

Publications, Inc., 1975.

[15] B. J. Verhaar, J. M. V. A. Koelman, H. T. C. Stoof, and 0. J. Luiten. Hyperfine

Contribution to Spin-Exchange Frequency Shifts in the Hydrogen Maser. Phys.
Rev. A, 35:3825, 1987.

[16] E. Tiesinga, B. J. Verhaar, H. T. C. Stoof, and D. van Bragt. Spin-Exchange

Frequency Shift in a Cesium Fountain. Phys. Rev. A, 45:R2671, 1992.

[17] K. Gibble and S. Chu. Laser-Cooled Cs Frequency Standard and a Measurement

of the Frequency Shift due to Ultracold Collisions. Phys. Rev. Lett., 70:1771,
1993.

[18] S. Ghezali, Ph. Laurent, S. N. Lea, and A. Clairon. An Experimental Study

of the Spin-Exchange Frequency Shift in a Laser-Cooled Cesium Fountain Fre-
quency Standard. Europhys. Lett., 36:25, 1996.

174
[19] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, and E. A.
Cornell. Observation of Bose-Einstein Condensation in a Dilute Atomic Vapor.
Science, 269:198, 1995.

[20] K. B. Davis, M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. S. Durfee,

D. M. Kurn, and W. Ketterle. Bose-Einstein Condensation in a Gas of Sodium
Atoms. Phys. Rev. Lett., 75:3969, 1995.

[21] C. C. Bradley, C. A. Sackett, J. J. Tollet, and R. G. Hulet. Evidence of Bose-

Einstein Condensation in an Atomic Gas with Attractive Interactions. Phys.
Rev. Lett., 75:1687, 1995.

[22] C. C. Bradley, C. A. Sackett, and R. G. Hulet. Bose-Einstein Condensation of

Lithium: Observation of Limited Condensate Number. Phys. Rev. Lett., 78:985,
1997.

[23] D. G. Fried. Bose-Einstein Condensation of Atomic Hydrogen. Phd, MIT, 1999.

[24] J. M. Doyle. Energy Distribution Measurements of Magnetically Trapped Spin

Polarized Atomic Hydrogen: Evaporative Cooling and Surface Sticking. Phd,
MIT, 1991.

[25] C. E. Hecht. The Possible Superfluid Behaviour of Hydrogen Atom Gases and
Liquids . Physica, 25:1159, 1959.

[26] W. C. Stwalley and L. H. Nosanow. Possible "New" Quantum Systems. Phys.

Rev. Lett., 36:910, 1976.

[27] R. D. Etters, J. V. Dugan, and R. W. Palmer. The Ground State Properties

of Spin-Aligned Atomic Hydrogen, Deuterium, and Tritium. J. Chem. Phys.,
62:313, 1975.

[28] W. Kolos and L. Wolniewicz. Potential Energy Curves for the XE+, b3F-+,
and C 1I1, States of the Hydrogen Molecule. J. Chem. Phys., 43:2429, 1965.

175
[29] W. Kolos and L. Wolniewicz. Variational Calculation of the Long-Range Inter-
action between two Ground-State Hydrogen Atoms. Chem. Phys. Lett., 24:457,
1974.

[30] H. A. Bethe and E. E. Salpeter. Quantum Mechanics of One- and Two-Electron

Atoms, chapter 3. Plenum Publishing Corporation, 1977.

[31] I. F. Silvera and J. T. M. Walraven. Phys. Rev. Lett., 21:164, 1980.

[32] Richard W. Cline, Thomas J. Greytak, and Daniel Kleppner. Nuclear Polar-
ization of Spin-Polarized Hydrogen. Phys. Rev. Lett., 47:1195, 1981.

[33] D. A. Bell, H. F. Hess, G. P. Kochanski, S. Buchman, L. Pollack, Y. M.

Xiao, D. Kleppner, and T. J. Greytak. Relaxation and Recombination in Spin-
polarized Atomic Hydrogen. Phys. Rev. B, 34:7670, 1986.

[34] Harald F. Hess. Evaporative Cooling of Magnetically Trapped and Compressed

Spin-Polarized Hydrogen. Phys. Rev. B, 34:3476, 1986.

[35] David E. Pritchard. Cooling Neutral Atoms in a Magnetic Trap for Precision
Spectroscopy. Phys. Rev. Lett., 51:1336, 1983.

[36] I. D. Setija, H. G. C. Werij, 0. J. Luiten, M. W. Reynolds, T. W. Hijmans, and

J. T. M. Walraven. Optical Cooling of Atomic Hydrogen in a Magnetic Trap
Phys. Rev. Lett., 70:2257, 1993.

[37] D. S. Zimmerman and A. J. Berlinsky. The Sticking Probability for Hydrogen

Atoms on the Surface of Liquid 4He. Can. J. Phys., 61:508, 1983.

[38] B. W. Statt, W. N. Hardy, A. J. Berlinsky, and E. Klein. J. Low Temp. Phys.,

61:471, 1985.

[39] J. Helffrich, M. Maley, M. Krusius, and J. C. Wheatley. Hydrogen Dissociation

Below 1 K. J. Low Temp. Phys., 66:277, 1987.

[40] H. F. Hess, G. P. Kochanski, J. M. Doyle, T. J. Greytak, and D. Kleppner.

Spin-Polarized Hydrogen Maser. Phys. Rev. A, 34:R1602, 1986.

176
[41] V. W. Macalpine and R. 0. Schildknecht. Coaxial Resonators with Helical Inner
Conductor. In Proc. IRE, page 2099, 1959.

[42] H. T. C. Stoof, J. M. V. A. Koelman, and B. J. Verhaar. Spin-Exchange

and Dipole Relaxation Rates in Atomic Hydrogen: Rigorous and Simplified
Calculations. Phys. Rev. B, 38:4688, 1988.

[43] H. F. Hess, G. P. Kochanski, J. M. Doyle, N. Masuhara, D. Kleppner, and T. J.

Greytak. Magnetic Trapping of Spin-Polarized Atomic Hydrogen. Phys. Rev.
Lett., 59:672, 1987.

[44] R. van Roijen, J. J. Berkhout, S. Jaakkola, and J. T. M. Walraven. Experiments

with Atomic Hydrogen in a Magnetic Trapping Field. Phys. Rev. Lett., 61:931,
1988.

[45] A. J. Berlinsky, W. N. Hardy, and B. W. Statt. Theory of Nuclear Spin-Lattice

Relaxation of Spin-Polarized Hydrogen on Liquid-Helium-Coated Surfaces due
to Magnetic Particles in the Substrate. Phys. Rev. B, 35:4831, 1987.

[46] D. G. Friend and R. D. Etters. A Dilute Hard-Sphere Bose-Gas Model Calcu-

lation of Low-Density Atomic-Hydrogen Gas Properties. J. Low Temp. Phys.,
39:409, 1980.

[47] W. Ketterle and N. J. van Druten. Evaporative Cooling of Trapped Atoms.

In B. Bederson and H. Walther, editors, Advances in Atomic, Molecular, and
Optical Physics, number 37, page 181, San Diego, 1996. Academic Press.

[48] I. A. Yu, J. M. Doyle, J. C. Sandberg, C. L. Cesar, D. Kleppner, and T. J.

Greytak. Evidence for Universal Quantum Reflection of Hydrogen from Liquid
4 He. Phys. Rev. Lett., 71:1589, 1993.

[49] G. M. Kavoulakis, C. J. Pethick, and H. Smith. Relaxation Processes in Clouds

of Trapped Bosons above the Bose-Einstein Condensation Temperature. Phys.
Rev. Lett., 81:4036, 1998.

177
[50] 0. J. Luiten, M. W. Reynolds, and J. T. M. Walraven. Kinetic Theory of the
Evaporative Cooling of a Trapped Gas. Phys. Rev. A, 53:381, 1996.

[51] N. Masuhara, J. M. Doyle, J. C. Sandberg, D. Kleppner, and T. J. Greytak.

Evaporative Cooling of Spin-Polarized Atomic Hydrogen. Phys. Rev. Lett.,
61:935, 1988.

[52] J. M. Doyle, J. C. Sandberg, I. A. Yu, C. L. Cesar, D. Kleppner, and T. J. Grey-

tak. Hydrogen in the Submillikelvin Regime: Sticking Probability on Superfuid
'He. Phys. Rev. Lett., 67:603, 1991.

[53] P. W. H. Pinske, A. Moske, M. Weidemiiller, M. W. Reynolds, T. W. Hijmans,

and J. T. M. Walraven. One-Dimensional Evaporative Cooling of Magnetically
Trapped Atomic Hydrogen. Phys. Rev. A, 57:4747, 1998.

[54] D. E. Pritchard, K. Helmerson, and A. G. Martin. In S. Haroche and J. C.

Gay and G. Grynberg, editor, Atomic Physics 11, page 179. World Scientific,
Singapore, 1989.

[55] I. A. Yu. Ultracold Surface Collisions: Sticking Probability of Atomic Hydrogen

on Superfluid 4He. Phd, MIT, 1993.

[56] J. M. Doyle, J. C. Sandberg, N. Masuhara, I. A. Yu, D. Kleppner, and T. J.

Greytak. Energy Distributions of Trapped Atomic Hydrogen. J. of the Opt.
Soc. Am. B, 6:2244, 1989.

[57] 0. J. Luiten, H. G. C. Werij, I. D. Setija, M. W. Reynolds, T. W. Hijmans, and

J. T. M. Walraven. Lyman-a Spectroscopy of Magnetically Trapped Atomic
Hydrogen. Phys. Rev. Lett., 70:544, 1993.

[58] P. W. H. Pinske, A. Moske, M. Weidemiiller, M. W. Reynolds, T. W. Hijmans,

J. T. M. Walraven, and C. Zimmerman. Resonance Enhanced Two-Photon
Spectroscopy of Magnetically Trapped Atomic Hydrogen. Phys. Rev. Lett.,
79:2423, 1997.

178
[59] J. C. Sandberg. Research Towards Laser Spectroscopy of Trapped Hydrogen.
Phd, MIT, 1993.

[60] C. L. Cesar. Two-Photon Spectroscopy of Trapped Atomic Hydrogen. Phd,

MIT, 1995.

[61] T. W. Hnsch, S. A. Lee, R. Wallenstein, and C. Wieman. Doppler-Free Two-

Photon Spectroscopy of Hydrogen 1S-2S*. Phys. Rev. Lett., 34:307, 1975.

[62] C. J. Foot, B. Couillaud, R. G. Beausoleil, and T. W. Hansch. Continuous-

Wave Two-Photon Spectroscopy of the 1S-2S Transition in Hydrogen. Phys.
Rev. Lett., 54:1913, 1985.

[63] C. Zimmerman, R. Kallenbach, and T. W. Hdnsch. High-Resolution Spec-

troscopy of the Hydrogen 1S-2S Transition in an Atomic Beam. Phys. Rev.
Lett., 65:571, 1990.

[64] B. Gross, A. Huber, M. Niering, M. Weitz, and T. W. Hdnsch. Optical Ramsey

Spectroscopy of Atomic Hydrogen. Europhys. Lett., 44:186, 1998.

[65] F. Schmidt-Kaler, D. Leibfried, M. Weitz, and T. W. Hdnsch. Precision Mea-

surement of the Isotope Shift of the 1S-2S Transition of Atomic Hydrogen and
Deuterium. Phys. Rev. Lett., 70:2261, 1993.

[66] J. H. Tung, X. M. Ye, G. J. Salamo, and F. T. Chan. Two-Photon Decay of

Hydrogenic Atoms. Phys. Rev. A, 30:1175, 1984.

[67] J. L. Wiza. Microchannel Plate Detectors. Nuclear Instruments and Methods,

162:587, 1979.

[68] Galileo Electro-Optics Corporation. Sturbridge, MA. (800)648-1800.

[69] Hamamatsu Corporation. Bridgewater, NJ. (908)231-0960.

[70] Acton Research Corporation. Acton, MA. (508)263-3584.

179
[71] R. Kallenbach, C. Zimmerman, D. H. McIntyre, and T. W. Hinsch. A blue dye
laser with sub-kilohertz stability. Opt. Comm., 70:56, 1989.

[72] A. E. Siegman. Lasers. University Science Books, 1986.

[73] R. W. P. Drever, J. L. Hall, F. V. Kowalski, J. Hough, G. M. Ford, A. J. Mun-

ley, and H. Ward. Laser Phase and Frequency Stabilization Using an Optical
Resonator. Appl. Phys. B, 31:97, 1983.

[74] C. Chen, B. Wu, A. Jiang, and G. You. A New-Type Ultraviolet SHG Crystal
- #-BaB 2 0 4. Scienta Sinica(Ser. B), 28:235, 1985.

[75] T. W. Hinsch and B. Couillaud. Laser Frequency Stabilization by Polarization

Spectroscopy of a Reflecting Reference Cavity. Opt. Comm., 35:441, 1980.

[76] C. Borde. Forme de Rai en Spectroscopie a Deux Quanta sans Elargissement

Doppler. C. R. Hebd. Sean. Acad. Sci. B, 282:341, 1976.

[77] F. Biraben, M. Bassini, and B. Cagnac. Line-Shapes in Doppler-Free Two-

Photon Spectroscopy. The Effect of Finite Transit Time. J. Phys. (Paris),
40:445, 1979.

[78] J. J. Sakurai. Modern Quantum Mechanics. Addison-Wesley Publishing Com-

pany, Inc., 1985.

[79] R. G. Beausoleil and T. W. Hinsch. Ultrahigh-Resolution Two-Photon Optical

Ramsey Spectroscopy of an Atomic Fountain. Phys. Rev. A, 33:1661, 1986.

[80] F. Bassani, J. J. Forney, and A. Quattropani. Choice of Gauge in Two-Photon

Transitions: ls-2s Transition in Atomic Hydrogen. Phys. Rev. Lett., 39:1070,
1977.

[81] R. H. Dicke. The Effect of Collisions upon the Doppler Width of Spectral Lines.
Phys. Rev., 89:472, 1953.

[82] A. Abragam. The Mdssbauer Effect. Gordon and Breach, 1964.

180
[83] A. Burgess. Tables of Hydrogenic Photoionization Cross-Sections and Recom-
binations Coefficients. Me. Royal Astron. Soc., 69:1, 1965.

[84] S. Klarsfield. Radiative Decay of Metastable Hydrogenic Atoms. Phys. Lett.,

30A:38, 1969.

[851 H. A. Bethe and E. E. Salpeter. Quantum Mechanics of One- and Two-Electron

Atoms, chapter 4. Plenum Publishing Corporation, 1977.

[86] N. F. Ramsey. Molecular Beams, chapter 5. Oxford University Press, 1956.

[87] D. J. Wineland, W. M. Itano, J. C. Bergquist, and R. G. Hulet. Laser-Cooling

Limits and Single-Ion Spectroscopy. Phys. Rev. A, 36:2220, 1987.

[88] M. Gatzke, G. Birkl, P. S. Jessen, A. Kastberg, S. L. Rolston, and W. D.

Phillips. Temperature and Localization of Atoms in Three-Dimensional Optical
Lattices. Phys. Rev. A, 55:R3987, 1997.

[89] M. J. Jamieson, A. Dalgarno, and M. Kimura. Scattering Lengths and Effective

Ranges for He-He and Spin-Polarized H-H and D-D Scattering. Phys. Rev. A,
51:2626, 1995.

[90] M. J. Jamieson, A. Dalgarno, and J. M. Doyle. Scattering Lengths for Collisions

of Ground State and Metastable State Hydrogen Atoms. Mol. Phys., 87:817,
1996.

[91] A. Dalgarno, 1998. Private communication.

[92] J. W. Liu and S. Hagstrom. J. Phys. B, 27:L729, 1994.

[93] W. Kolos and J. Rychlewski. J. Mol. Phys., 143:237, 1990.

[941 P. S. Julienne and F. H. Mies. Collisions of Ultracold Atoms. J. Opt. Soc. Am.
B, 6:2257, 1989.

[95] K. Huang. Statistical Mechanics, chapter 10. John Wiley and Sons, 1963.

181
 [96] R. K. Pathria. Statistical Mechanics, chapter 10. Pergamon Press, 1972.

[97] M.-O. Mewes, M. R. Andrews, D. M. Kurn, D. S. Durfee, C. G. Townsend, and

W. Ketterle. Output Coupler for Bose-Einstein Condensed Atoms. Phys. Rev.
Lett., 78:582, 1997.

[98] G. B. Baym. Lectures on Quantum Mechanics, chapter 18. Addison-Wesley

Publishing Company, 1990.

[99] A. Dalgarno, 1998. Preliminary calculations indicate this effect is small. Private
communication.

[100] L. Van Hove. Correlations in Space and Time and Born Approximation Scat-
tering in Systems of Interacting Particles . Phys. Rev., 95:249, 1954.

[101] W. Ketterle and H.-J. Miesner. Coherence Properties of Bose-Einstein Conden-

sates and Atom Lasers. Phys. Rev. A, 56:3291, 1997.

[102] V. L. Ginzburg and L. P. Pitaevskii. Sov. Phys. JETP, 7:858, 1958.

[103] E. P. Gross. J. Math. Phys., 4:195, 1963.

[104] G. Baym and C. J. Pethick. Ground State Properties of Magnetically Trapped

Bose-Condensed Rubidium Gas. Phys. Rev. Lett., 76:6, 1996.

[105] M. R. Mathews, D. S. Hall, D. S. Jin, J. R. Ensher, C. E. Wieman, E. A.

Cornell, F. Dalfovo, C. Minniti, and S. Stringari. Dynamical Response of a
Bose-Einstein Condensate to a Discontinuous Change in Internal State. Phys.
Rev. Lett., 81:243, 1998.

[106] D. S. Hall, M. R. Mathews, J. R. Ensher, C. E. Wieman, and E. A. Cornell.

Dynamics of Component Separation in a Binary Mixture of Bose-Einstein Con-
densates. Phys. Rev. Lett., 81:1539, 1998.

[107] 1998. Recent, unpublished work from the laboratory of W. Ketterle at MIT.

182
[108] 1998. Recent, unpublished work from the laboratory of W. Phillips at NIST,
Gaithersburg. Reported at the DAMOP meeting of the APS.

[109] L. Levitov, 1998. Private communication.

[110] K. Huang. Statistical Mechanics, page 182. John Wiley and Sons, 1963.

[111] H. A. Bethe and E. E. Salpeter. Quantum Mechanics of One- and Two-Electron

Atoms, page 256. Plenum Publishing Corporation, 1977.

[112] J. J. Sakurai. Modern Quantum Mechanics. Addison-Wesley Publishing Com-

pany, Inc., 1985.

[113] V. Bagnato, D. Pritchard, and D. Kleppner. Bose-Einstein Condensation in an

External Potential. Phys. Rev. A, 35:4354, 1987.

[114] J. R. Ensher, D. S. Jin, M. R. Mathews, C. E. Wieman, and E. A. Cornell. Bose-

Einstein Condensation in a Dilute Gas: Measurement of Energy and Ground-
State Occupation. Phys. Rev. Lett., 77:4984, 1996.

[115] T. W. Hijmans, Yu. Kagan, G. V. Shlyapnikov, and J. T. M. Walraven. Bose

Condensation and Relaxation in Magnetically Trapped Atomic Hydrogen. Phys.
Rev. A, 48:12886, 1993.

[116] J. M. Vogels, C. C. Tsai, R. S. Freeland, S. J. J. M. F. Kokkelmans, B. J.

Verhaar, and D. J. Heinzen. Prediction of Feshbach Resonances in Collisions of
Ultracold Rubidium Atoms. Phys. Rev. A, 56:R1067, 1997.

[117] E. Tiesinga, C. J. Williams, P. S. Julienne, K. M. Jones, P. D. Lett, and W. D.

Phillips. J. Res. Natl. Inst. Stand. Technol., 101:505, 1996.

[118] M. R. Andrews, C. G. Townsend, H.-J. Miesner, D. S. Durfee, D. M. Kurn,

and W. Ketterle. Observation of Interference Between Two Bose Condensates.
Science, 275:637, 1997.

183
[119] M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. M. Kurn, D. S. Durfee, and
W. Ketterle. Bose-Einstein Condensation in a Tightly Confining dc Magnetic
Trap. Phys. Rev. Lett., 78:582, 1997.

[120] M. J. Holland, D. S. Jin, M. L. Chiofalo, and J. Cooper. Emergence of In-

teractions Effects in Bose-Einstein Condensation. Phys. Rev. Lett., 78:3801,
1997.

[121] M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. M. Kurn, D. S. Durfee,

C. G. Townsend, and W. Ketterle. Collective Excitations of a Bose-Einstein
Condensate in a Magnetic Trap. Phys. Rev. Lett., 77:988, 1996.

[122] D. S. Jin, J. R. Ensher, M. R. Mathews, C. E. Wieman, and E. A. Cornell.

Collective Excitation of a Bose-Einstein Condensate in a Dilute Gas. Phys.
Rev. Lett., 77:420, 1996.

[123] M. R. Andrews, D. M. Kurn, H.-J. Miesner, D. S. Durfee, C. G. Townsend,

S. Inouye, and W. Ketterle. Propogation of Sound in a Bose-Einstein Conden-
sate. Phys. Rev. Lett., 79:553, 1997.

[124] J. Stenger, S. Inouye, D. M. Stamper-Kurn, H.-J. Miesner, A.P. Chikkatur, and

W. Ketterle. Spin Domains in Ground-State Spinor Bose-Einstein Condensates.
Nature, 396:345, 1998.

[125] E. A. Burt, R. W. Ghrist, C. J. Myatt, M. J. Holland, E. A. Cornell, and C. E.

Wieman. Coherence, Correlations, and Collisions: What one Learns about
Bose-Einstein Condensates from Their Decay. Phys. Rev. Lett., 79:337, 1997.

[126] H.-J. Miesner, D. M. Stamper-Kurn, M. R. Andrews, D. S. Durfee, S. Inouye,

and W. Ketterle. Bosonic Stimulation in the Formation of a Bose-Einstein
Condensate. Science, 279:1005, 1998.

[127] Yu. M. Kagan, B. V. Svistunov, and G. V. Shlyapnikov. Kinetics of Bose

Condensation in an Interacting Bose Gas. Sov. Phys. JETP, 75:387, 1992.

184
[128] S. Giorgini, L. P. Pitaevskii, and S. Stringari. Anomalous Fluctuations of the
Condensate in Interacting Bose Gases. Phys. Rev. Lett., 80:5040, 1998.

[129] J. M. Vaughn. The Fabry-PerotInterferometer: History, Theory, Practice and

Applications. IOP Publishing, Ltd, 1989.

[130] J. C. Bergquist and D. Berkland, 1998. DAMOP meeting of the APS.

[131] S. Seel, R. Storz, G. Ruoso, J. Mlynek, and S. Schiller. Cryogenic Optical

Resonators: A New Tool for Laser Frequency Stabilization at the 1 Hz Level.
Phys. Rev. lett., 78:4741, 1997.

[132] K. Huang. Statistical Mechanics, chapter 3. John Wiley and Sons, 1963.

[133] D. H. McIntyre, W. M. Fairbank, Jr., S. A. Lee, T. W. Hinsch, and E. Riis.

Interferometric Frequency Measurement of 130 Te Reference Transitions at 486
2

nm. Phys. Rev. A, 41:4632, 1990.

[134] J. D. Gillaspy and C. J. Sansonetti. Absolute Wavelength Determinations in

Molecular Tellurium: New Reference Lines for Precision Laser Spectroscopy. J.
Opt. Soc. Am. B, 8:2414, 1991.

[135] D. H. McIntyre. High Resolution Spectroscopy of Tellurium and Hydrogen: A

measurement of the Rydberg Constant. Phd, Stanford University, 1987.

[136] J. Cariou and P. Luc. Atlas du Spectre d'Absorption de la Molecule de Tellure.

Laboratoire Aime Cotton C.N.R.S., Orsay, France, 1980.

[137] S. Haroche and F. Hartman. Theory of Saturated-Absorption Line Shapes.

Phys. Rev. A, 6:1280, 1972.

185